CHEP 07

Europe/Zurich

,
Description
Computing in High Energy and Nuclear Physics
Support
• Monday, September 3
• Poster 1: Day 1
• 1
Anvil' a case study of the Unified Software Development Process
The Unified Software Development Process (USDP) defines a process for developing software from the initial inception to the final delivery. The process creates a number of difference models of the final deliverable; the use case, analysis, design, deployment, implementation and test models. These models are developed using an iterative approach that breaks down into four main phases; inception, elaboration, construction and transition. The 'anvil' project is required to produce the Experiment Control sub-system for the IceCube Neutrino Telescope based at the South Pole. This project used the USDP as the basis of its development. It turned out to be small enough and self-contained enough to act as a valuable case study of the application of the USDP in HEP software. This paper shows how the various USDP models evolved during the development of the final deliverable. It also demonstrates the various phases of development and how these were used to mitigate risk. Finally, this paper discusses the final set of artifacts created by the USDP and presents a number of templates to help with their creation in future projects.
Speaker: Dr Simon Patton (LAWRENCE BERKELEY NATIONAL LABORATORY)
• 2
A distributed approach for a regional grid operation centre
Forschungszentrum Karlsruhe is one of the largest science and engineering research institutions in Europe. The resource centre GridKa as part of this science centre is building up a Tier 1 centre for the LHC project. Embedded in the European grid initiative EGEE, GridKa also manages the ROC (regional operation centre) for the German Swiss region. A ROC is responsible for regional coordination, organisation and support of the sustainable operations grid infrastructure in EGEE. A particularity of this specific ROC is its distributed organisational structure. The German Swiss operations community, currently consisting of 14 resource centres, is supported and organised by a team of 6 partners. On the one hand a consequence of this decentral approach is a larger effort for organisation and management. On the other hand grid knowledge and expertise is naturally spread among partners with various backgrounds. As a consequence grid technology is being developed and deployed with a wider usability over more use cases and science communities, then it would be in a central approach. Different aspects of this regional specific organisation structures as part of the global EGEE grid operations as well as roadmap for upcoming ROC services and tasks will be highlighted. The successful integration of grid tools for operations and support leads to a sustainable structure of the German Swiss region within a world wide grid community.
Speaker: Dr Sven Hermann (Forschungszentrum Karlsruhe)
• 3
A Light-weight Intrusion Detection System for clusters
The RPMVerify package is a light weight intrusion detection system (IDS) which is used at CERN as part of the wider security infrastructure. The package provides information about potentially nefarious changes to software which has been deployed using the RedHat Package Management system (RPM). The purpose of the RPMVerify project has been to produce a system which makes use of the existing CERN infrastructure and tackles the scalability limitations of existing IDSs. In this paper we discuss its design, implementation, limitations, and our experiences in using it. We will specifically comment from the system administration and service management perspective.
Speaker: Alasdair Earl (CERN)
• 4
A Scientific Overview of Network Connectivity and Grid Infrastructure in South Asian Countries
The future of Computing in High Energy Physics (HEP) applications depends on both the Network and Grid infrastructure. Some South Asian countries such as India and Pakistan are making progress in this direction by not only building Grid clusters, but also by improving their network infrastructure. However to facilitate the use of these resources, they need to overcome the issues of network connectivity to be among the leading participants in Computing for HEP experiments. In this paper we classify the connectivity for academic and research institutions of South Asia. The quantitative measurements are carried out using the PingER methodology; an approach that induces minimal ICMP traffic to gather end-to-end network statistics. The PingER project has been measuring the Internet performance for the last decade. Currently the measurement infrastructure comprises of over 700 hosts in more than 130 countries which collectively represents approximately 99% of the world's Internet-connected population. Thus, we are well positioned to characterize the world's connectivity. Here we present the current state of the National Research and Educational Networks (NRENs) and Grid Infrastructure in the South Asian countries and identify the areas of concern. We also present comparisons between South Asia and other developing as well as developed regions. We show that there is a strong correlation between the Network performance and several Human Development indices.
Speaker: Mr Shahryar Khan (Stanford Linear Acclerator Center)
• 5
A Search Engine for the Engineering and Equipment Data Management System (EDMS) at CERN
CERN, the European Laboratory for Particle Physics, located in Geneva - Switzerland, is currently building the LHC, a 27 km particle accelerator. The equipment life-cycle management of this project is provided by the Engineering and Equipment Data Management System (EDMS) Service. Using Oracle, it supports the management and follow-up of different kinds of documentation through the whole life cycle of the LHC project: design, manufacturing, installation, commissioning data etc... The equipment data collection phase is now slowing down and the project is getting closer to the “As-Built” phase; the phase of the project consuming and exploring the large volumes of data stored since 1996. Searching through millions of pieces of information (documents, pieces of equipment, operations...) multiplied by dozens of points of view (operators, maintainers...) require an efficient and flexible search engine. This paper describes the process followed by the team to implement the search engine for LHC As-built project in the EDMS Service. The emphasis is put on the design decision to decouple the search engine from any user interface, potentially enabling other systems to also use it. Projections, algorithms, and the planned implementation are described in this paper. The implementation of the first version started in early 2007.
Speaker: Mr Andrey Tsyganov (Moscow Physical Engineering Inst. (MePhI))
• 6
A study of the accuracy of Network Time Protocol client synchronization in large computing clusters
As computing systems become more distributed and as networks increase in throughput and resources become ever increasingly dispersed over multiple administrative domains, even continents, there is a greater need to know the performance limits of the underlying protocols which make the foundations of complex computing and networking architectures. One such protocol is the Network Time Protocol (NTP) which is often overlooked as an important part of any large scale computing system. With the adoption of new highly distributed technologies, such as those employed in grid computing, the increasing number of users and resources will test not only the synchronization of these resources but also the transaction logging and event correlation in any problem resolution/diagnostic systems. In essence, good quality and reliable time synchronization is a key component to the actual operation of any large scale production system incorporating many components. In this paper we present the CERN NTP server and client architecture and discuss the statistical quality of time synchronization of 4 computing clusters of increasing size from approximately 50 to 3000 nodes and inter-connected via a high-performance 10Gbit/s symmetrically routed network backbone infrastructure. Each cluster is dedicated to a specific task or application resulting in various IO load profiles, some more deterministic than others. The relationship between the reliability of time synchronization, system load and network IO is analysed and optimization suggestions are presented.
Speaker: Dr Nick Garfield (CERN)
• 7
ALICE DAQ Online Transient Data Storage
ALICE is a dedicated heavy-ion detector to exploit the physics potential of nucleus-nucleus (lead-lead) interactions at LHC energies. The aim is to study the physics of strongly interacting matter at extreme energy densities, where the formation of a new phase of matter, the quark-gluon plasma, is expected. Running in heavy-ion mode the data rate from event building to permanent storage is expected to be around 1.25 GB/s. To continue data recording even in the event of hardware failure or connection problems, a large disk pool has been installed at the experiment's site as buffering layer between the DAQ and the remote (~5km) tape facility in the CERN Computing Centre. This Transient Data Storage (TDS) disk pool has to provide the bandwidth to be able to simultaneously absorb data from the event building machines and to move data to the tape facility. The aggregated bandwidth of the TDS is expected to exceed 3 GB/s in mixed I/O traffic. Extensive tests have been carried out on various hardware and software solutions with the goal to build a common file space shared by ~60 clients, whilst still providing maximum bandwidth per client (~400MB/s, 4Gbps Fibre Channel), fail-over safety and redundancy. This talk will present the chosen hardware and software solution, the configuration of the TDS pool and the various modes of operation in the ALICE DAQ framework. It will also present the results of the performance tests carried out during the last ALICE Data Challenge.
Speaker: Mr Ulrich Fuchs (CERN & Ludwig-Maximilians-Universitat Munchen)
• 8
Alignment data streams for the ATLAS Inner Detector.
The ATLAS experiment uses a complex trigger strategy to be able to achieve the necessary Event Filter rate output, making possible to optimize the storage and processing needs of these data. These needs are described in the ATLAS Computing Model which embraces Grid concepts. The output coming from the Event Filter will consist of four main streams: the physical stream, express stream, calibration stream, and a diagnostic stream. The calibration stream will be transferred to the Tier-0 facilities which will provide the prompt reconstruction of this stream with a minimum latency of 8 hours, producing calibration constants of sufficient quality to permit a first-pass processing. The Inner Detector community is developing and testing an independent common calibration stream selected at the Event Filter after track reconstruction. It is composed of raw data, in byte-stream format, contained in ROB's with hit information of the selected tracks, and it will be used to derive and update a set of calibration and alignment constants after every fill. This option was selected because makes use of the Byte Stream Converter infrastructure and possibly give us a better bandwidth usage and storage capability's. Processing is done using specialized algorithms running in Athena framework in dedicated Tier-0 resources, and the alignment constants will be stored and distributed using the COOL conditions database infrastructure. The work is addressing in particular the alignment requirements, the needs for track and hit selection and the timing issues.
Speaker: Mr Belmiro Antonio Venda Pinto (Faculdade de Ciencias - Universidade de Lisboa)
• 9
An Inconvenient Truth: file-level metadata and in-file metadata caching in the (file-agnostic) ATLAS distributed event store
In the ATLAS event store, files are sometimes "an inconvenient truth." From the point of view of the ATLAS distributed data management system, files are too small--datasets are the units of interest. From the point of view of the ATLAS event store architecture, files are simply a physical clustering optimization: the units of interest are event collections-- sets of events that satisfy common conditions or selection predicates-- and such collections may or may not have been accumulated into files that contain those events and no others. It is nonetheless important to maintain file-level metadata, and to cache metadata in event data files. When such metadata may or may not be present in files, or when values may have been updated after files are written and replicated, a clear and transparent model for metadata retrieval from the file itself or from remote databases is required. In this paper we describe how ATLAS reconciles its file and non-file paradigms, the machinery for associating metadata with files and event collections, and the infrastructure for metadata propagation from input to output for provenance record management and related purposes.
Speaker: Dr David Malon (Argonne National Laboratory)
• 10
An interface to tape for disk pool managers
The disk pool managers in use in the HEP community focus on managing disk storage but at the same time rely on a mass storage i.e. tape based system either to offload data that has not been touched for a long time or for archival purposes. Traditionally tape handling systems like HPSS by IBM or Enstore developed at FNAL are used because they offer specialized features to overcome the limitations of the sequential data access of tape. Not all centers have the resources to support special purpose tape handling systems but in many environments like FZK, Tivoli Storage Manager is in use for regular desktop backups. The paper describes the dCache to TSM interface TSS, that has been developed and is in use for over a year at FZK/GridKa. It served during the last WLCG services challenges and the recent experiment data challenges with peak rates of 300 MB/s into 8 tape drives. TSS and TSM make use of SAN connected data movers that write in parallel to tape. The TSS interface can be used for dCache and xrootd and offers a queuing layer between the disk cache and the tape backend in order to enhance store and pre-staging operations. The current status of the project and achieved data rates as well as future enhancements are presented.
Speaker: Jos Van Wezel (Forschungszentrum Karlsruhe (FZK/GridKa))
• 11
ATLAS Computing System Commissioning - Simulation Production Experience
During 2006-07, the ATLAS experiment at the Large Hadron Collider launched a massive Monte Carlo simulation production exercise to commission software and computing systems in preparation for data in 2007. In this talk, we will describe the goals and objectives of this exercise, the software systems used, and the tiered computing infrastructure deployed worldwide. More than half a petabyte of data was generated at more than 50 different sites. The results of this year-long exercise will be summarized, with special emphasis on the lessons learned from an international distributed computing exercise of unprecedented size and scope.
Speaker: Kaushik De (UT-Arlington)
• 12
ATLAS Conditions Database Experience with the LCG COOL Conditions Database Project
One of the most challenging task faced by the LHC experiments will be the storage of "non-event data" produced by calibration and alignment stream processes into the Conditions Database. For the handling of this complex experiment conditions data the LCG Conditions Database Project has implemented COOL, a new software product designed to minimise the duplication of effort by developing a single implementation to support persistency for several relational technologies (Oracle,MySQL and SQLite). After several production releases of the COOL software, the project has be moved into the deployment phase in Atlas and LHCb, the two experiments that are developing the software in collaboration with CERN IT. In particular, the ATLAS Conditions Database, accessed by the offline reconstruction framework (ATHENA), is implemented using this COOL technology. The objects, stored or referenced in the COOL tables, have an associated start and end time (run or event number or absolute time-stamp) between they are valid (Interval of Validity). The storage and retrieving of data during a recontruction job inside ATHENA is guaranteed by the IOVService, a software interface between the COOL DB and the reconstruction algorithms. This work describes some practical examples already successfully tested in ATLAS and the further extensively tests of the entire chain foreseen during the next Computing and Detector Commissioning in 2007.
Speaker: Dr Monica Verducci (European Organization for Nuclear Research (CERN))
• 13
ATLAS Detector Maintenance And Operation Management Tool
The maintenance and operation of the ATLAS detector will involve thousands of contributors from 170 physics institutes. Planning and coordinating the action of ATLAS members, ensuring their expertise is properly leveraged and that no parts of the detector are under or overstaffed will be a challenging task. The ATLAS Maintenance and Operation (ATLAS M&O) application offers a fluent web based interface that combines the flexibility and comfort of a desktop application, intuitive data visualization and navigation techniques, with a lightweight service oriented architecture. We will review the application, its usage within the ATLAS experiment, its underlying design and the software project management techniques employed to complete the project.
Speaker: Mr Brice Copy (CERN)
• 14
ATLAS Tile Calorimeter on-line monitoring system based on the Event Filter
ATLAS Tile Calorimeter detector (TileCal) is presently involved in an intense phase of commissioning with cosmic rays and subsystems integration. Various monitoring programs have been developed at different level of the data flow to tune the set-up of the detector running conditions and to provide a fast and reliable assessment of the data quality. The presentation will focus on the on-line monitoring tools employed during TileCal detector commissioning and integration tests with cosmic rays and in particular on the monitoring system integrated in the highest level of the Trigger, the Event Filter (EF). The key feature of EF monitoring is the capability of performing detector and data quality control on the complete physics event at the trigger level, hence before it is stored on disk. In the on-line data flow, this is the only monitoring system in ATLAS capable of giving a comprehensive event quality feedback. The presentation will also show some monitoring results of the integration tests with other sub-detectors and performances and future upgrades of the current implementation.
Speakers: Nils Gollub (CERN) , Nils Gollub (University of Uppsala)
• 15
ATLAS's EventView Analysis Framework
The EventView Analysis Framework is currently the basis for much of the analysis software employed by various ATLAS physics groups (for example the Top, SUSY, Higgs, and Exotics working groups). In ATLAS's central data preparation, this framework provides an assessment of data quality and the first analysis of physics data for the whole collaboration. An EventView is a self-consistent interpretation of a physics event or equivalently the state of a specific analysis. Analyses are constructed at runtime by chaining and configuring modular components consisting of tools, C++ implementation of specific analysis algorithms, and modules, python grouping and configuration of various tool. A large common library of general tools and modules serve as the building blocks of nearly all of the steps of any analysis. The output is multiple simultaneous EventViews of every event, typically reflecting different choices of selections, reconstruction algorithms, combinatoric assignments, or input data (eg full or fast reconstruction or truth).
Speaker: Dr Amir Farbin (European Organization for Nuclear Research (CERN))
• 16
Beyond Grid Security
While many fields relevant to Grid security are already covered by existing working groups, their remit rarely goes beyond the scope of the Grid infrastructure itself. However, security issues pertaining to the internal set-up of compute centres have at least as much impact on Grid security. Thus, this talk will present briefly the EU ISSeG project (Integrated Site Security for Grids). To complement groups such as OSCT (Operational Security Coordination Team) and JSPG (Joint Security Policy Group), the purpose of ISSeG is to provide a holistic approach to security for Grid computer centres, from strategic considerations to an implementation plan and its deployment. The generalised methodology of Integrated Site Security (ISS) is based on the knowledge gained during its implementation at several sites as well as through security audits, and this will be briefly discussed. Several examples of ISS implementation tasks at the Forschungszentrum Karlsruhe will be presented, including segregation of the network for administration and maintenance and the implementation of Application Gateways. Furthermore, the web-based ISSeG training material will be introduced. This aims to offer ISS implementation guidance to other Grid installations in order to help avoid common pitfalls.
Speaker: Mr Bruno Hoeft (Forschungszentrum Karlsruhe)
• 17
Building a production-grade Grid infrastructure at DESY
As a partner of the international EGEE project in the German/Switzerland federation (DECH) and as a member of the national D-GRID initiative, DESY operates a large-scale production-grade Grid infrastructure with hundreds of CPU cores and hundreds of Terabytes of disk storage. As Tier-2/3 center for ATLAS and CMS DESY plays a leading role in Grid computing in Germany. DESY strongly support non-LHC VOs and fosters the Grid usage in other eScience fields. The DESY Grid infrastructure is the home of a number of global, regional, and local VOs, among them are the HERA experiments H1 and ZEUS ('hone', 'zeus'), the ILC community ('calice', 'ilc'), and for the International Lattice Data Grid ('ildg'). All of them are heavily using the Grid for event simulation and data analysis. The DESY Grid infrastructure incorporates all necessary Grid services to host its VOs and provides computing and storage resources for all supported VOs in the same Grid infrastructure. Main emphasis has been put on embedding the Grid infrastructure seamlessly in the DESY computer center. Crucial aspects are the choice of batch system and storage technologies (dCache). Important aspects are reliability of services by providing redundancies, monitoring, and administrative tools as well as scalable installation and updating procedures. In the contribution the CHEP07 we will describe the Grid set-up at DESY and discuss in detail concepts and implementation to achieve a scalable, pervasive, and easy to maintain system which meets the requirements of a global Grid.
Speaker: Dr Andreas Gellrich (DESY)
• 18
Circuit Oriented End-to-End Services for HEP Data Transfers
Most of today's data networks are a mixture of packet switched and circuit switched technologies, with Ethernet/IP on the campus and in data centers, and SONET/SDH over the wide area infrastructure. SONET/SDH allows creating dedicated circuits with bandwidth guarantees along the path, suitable for the use of aggressive transport protocols optimised for fast data transfer and without fairness constraints. On the downside, a provisioned, but under-utilised circuit may result in poor overall network utilisation, as the reserved bandwidth cannot be used by another flow. Addressing this issue, Virtual Concatenation (VCAT) and Link Capacity Adjustment Scheme (LCAS) are recent addition to SONET/SDH, and allow dynamic creation and hitless bandwidth adjustment of virtual circuits. Caltech and CERN have deployed optical multiplexing equipment supporting VCAT/LCAS on their US LHCNet transatlantic network, and agent-based grid and network monitoring software based on Caltech's MonALISA system, to provide on-demand end-to-end bandwidth guarantees for data transfers between Tier-N centres, following the GLIF concept for control plane interaction between the participating networks. This is being coordinated with the LHC experiments' management software for dataset distribution, and with the circuit segment-provisioning developments of ESnet, Internet2, Fermilab, BNL, GEANT2 and collaborators to form the end-to-end network paths. MonALISA is used to oversee the progress, troubleshoot and mitigate problems associated with dynamic provisioning in response to multiple transfer requests. We present our experience with operating VCAT/LCAS enabled network for transatlantic connections, along with the details of a first implementation of circuit oriented end-to-end services for data transfers between data centres.
Speaker: Artur Barczyk (Caltech)
• 19
CMS Software Deployment on the Open Science Grid (OSG)
The CMS experiment will begin data collection at the end of 2007 and released its software with new framework since the end of 2005. The CMS experiment employs a tiered distributed computing based on the Grids, the LHC Computing Grid (LCG) and the Open Science Grid (OSG). There are approximately 37 tiered CMS centers around the world. The number of the CMS software releases was three per month on average. This corresponds to roughly 100 CMS software installations. A set of software deployment tools has been developed for the installation, verification, and deprecation of a release. The tools are mainly targeted for deployment on the OSG. The main features are the capability of the instant release deployment and the corrective resubmission of the installation jobs. We also use an independent web-based deployment portal with the Grid security infrastructure login mechanism. We have been deploying approximately 220 CMS softwares releases and one gLite User Interface at each site using the tools on the OSG to provide LCG-OSG interoperability. We found the tools are reliable and can be adaptable to cope with problems with changes in the Grid computing environment and the software releases. We will expand the set of sites at which we maintain CMS software deployments on the OSG We will present the design of the tools, an analysis of statistics that we gathered during the operation of the tools, and our experience with the CMS software deployment on the OSG Grid computing environment.
Speaker: Dr Bockjoo Kim Kim (University of Florida)
• 20
CMS Tier0 - design, implementation and first experiences
With the upcoming LHC engineering run in November, the CMS Tier0 computing effort will be the one of the most important activities of the experiment. The CMS Tier0 is responsible for all data handling and processing of real data events in the first period of their life, from when the data is written by the DAQ system to a disk buffer at the CMS experiment site to when it is transferred from CERN to the Tier1 computer centers. The CMS Tier0 accomplishes three principle processing tasks: the realization of the data streaming model of CMS, the automated production of calibration and alignment constants and the first full reconstruction of the raw data. The presentation will describe the data streaming model of CMS and how this is implemented in the CMS trigger/DAQ and the Tier0. For the Tier0 this implementation means a reorganization of the data from a format determined by the demands of the CMS trigger/DAQ to a format that is determined by physics demands. We will also describe the design and implementation of the Prompt Calibration and Prompt Reconstruction workflows. The data flow underlying these workflows will be shown and first results from data challenges and scale tests will be presented.
Speaker: Dirk Hufnagel (for the CMS Offline/Computing group)
• 21
cMsg - A general purpose, publish-subscribe, interprocess communication implementation and framework
cMsg is software used to send and receive messages in the Jefferson Lab online and runcontrol systems. It was created to replace the several IPC software packages in use with a single API. cMsg is asynchronous in nature, running a callback for each message received. However, it also includes synchronous routines for convenience. On the framework level, cMsg is a thin API layer in Java, C, or C++ that can be used to wrap most message-based interprocess communication protocols. The top layer of cMsg uses this same API and multiplexes user calls to one of many such wrapped protocols (or domains) based on a URL-like string which we call a Uniform Domain Locator or UDL. One such domain is a complete implementation of a publish-subscribe messaging system using network communications and written in Java (user APIs in C and C++ too). This domain is built in a way which allows it to be used as a proxy server to other domains (protocols). Performance is excellent allowing the system not only to be used for messaging but also as a data distribution system.
Speaker: Dr Carl Timmer (TJNAF)
• 22
Commissioning with cosmic rays of the Muon Spectrometer of the ATLAS experiment at the Large Hadron Collider
The Muon Spectrometer of the ATLAS experiment is made of a large toroidal magnet, arrays of high-pressure drift tubes for precise tracking and dedicated fast detectors for the first-level trigger. All the detectors in the barrel toroid have been installed and commissioning has started with cosmic rays. These detectors are arranged in three concentric rings and the total area is about 7000 square meters. During the installation and commissioning of the detectors, data are usually taken with the magnet off, but a dedicated run took place with the magnetic field of the barrel toroid turned on. We present the procedure to control the response of the single detectors installed in the barrel toroid, Monitored Drift Tubes and Resistive Plate Chambers, and results of the first tests done with cosmic rays triggered by the first-level processor and read-out trough the ATLAS data acquisition. A comparison of the detector performance in magnetic field on and off will be presented together with a measurement of the cosmic ray flux in the underground experimental area. Details on the installation and commissioning schedule will be given in view of the completion of the instrumentation of the muon spectrometer for the first period of data taking with proton-proton collisions.
Speaker: Rosy Nikolaidou (DAPNIA)
• 23
Computationally efficient algorithms for the two-dimensional Kolmogorov-Smirnov test
Goodness-of-fit statistics measure the compatibility of random samples against some theoretical probability distribution function. The classical one-dimensional Kolmogorov-Smirnov test is a non-parametric statistic for comparing two empirical distributions, which defines the largest absolute difference between the two cumulative probability distribution functions as a measure of disagreement. Adapting this test to more than one dimension is a challenge because there are 2k-1 independent ways of defining a cumulative probability distribution function when k dimensions are involved. However there are many applications in experimental physics where comparing two-dimensional data sets is important. In this paper we discuss Peacock's version [3] of the Kolmogorov-Smirnov test for two-dimensional data sets which computes the differences between cumulative probability distribution functions in 4n2 quadrants and runs in O(n^3), for a sample of size n. We also discuss Fasano and Franceschini's variation [2] of Peacock's test that runs in O(n^2), Cooke's algorithm [1]for Peacock's test and ROOT's version of the two-dimensional Kolmogorov-Smirnov. We establish a lower-bound limit on the work for computing Peacock's test in W(n2lgn), which contradicts Cooke's claim that it possible to perform this test with an O(nlgn) algorithm. We also establish a lower-bound limit of W(nlgn) for Fasano and Franceschini's test, and present an optimal sequential algorithm for it. We finally discuss experimental results comparing two parallel algorithms implementing Peacock's test and an optimal algorithm for Fasano and Franceschini's test, and contrast these with the algorithm in ROOT which is based on the calculation of a mean of two one-dimensional Kolmogorov-Smirnov tests. References [1] A. Cooke. The muac algorithm for solving 2d ks test. http://www.acooke.org/jara/muac/algorithm.html. [2] G. Fasano and A. Franceschini. A multidimensional of the Kolmogorov-Smirnov test. Monthly Notices Royal Astronomy Society, 225:155-170, 1987. [3] J. A. Peacock. Two-dimensional goodness-of-fit testing in astronomy. Monthly Notices Royal Astronomy Society, 202:615-627, 1983.
Speaker: Dr Ivan D. Reid (School of Design and Engineering - Brunel University, UK)
• 24
Computing in IceCube
The IceCube neutrino telescope is a cubic kilometer Cherenkov detector currently under construction in the deep ice at the geographic South Pole. As of 2007, it has reached more than 25 % of its final instrumented volume and is actively taking data. We will briefly describe the design and current status, as well as the physics goals of the detector. The main focus will, however, be on the unique computing structure of the experiment, due to the remoteness of the detector and the limited access to satellites for data transfer. We will also outline the structure of the software framework used for filtering, reconstruction and simulation.
Speaker: Mr Georges Kohnen (Université de Mons-hainaut)
• 25
Construct an LHC scale national analysis facility
Based on todays understanding of LHC scale analysis requirements and the clear dominance of fast and high capacity random access storage, this talk will present a generic architecture for a national facility based on existing components from various computing domains. The following key areas will be discussed in detail and solutions will be proposed, building the overall architecture. 1. large scale cluster filesystems known in the HPC community. This community has shown the mature character of this technologies during the last months and evidently this success could be made usable for the demanding HEP community. 2. methods and systems for decentralized/delegated user administration. This also includes the integration with VOMS to allow seamless management with existing system. 3. data access to/from T2 (generally TierX) facilities by using SRM enabled cluster filesystems and high speed data access to local TierX storage resources if available. 4. batch process integration presenting exactly the same interface/behavious to users with resepect to their interactive ones. 5. accounting and monitoring system to enable true sharing of the single resource between all participating VOs. 6. criteria showing the scaling properties in the PB data region and more than 1000 CPUs.
Speaker: Mr Martin Gasthuber (Deutsches Elektronen Synchrotron (DESY))
• 26
Control and monitoring of alignment data for the ATLAS endcap Muon Spectrometer at the LHC
The ATLAS Muon Spectrometer is constructed out of 1200 drift tube chambers with a total area of nearly 7000 square meters. It must determine muon track positions to a very high precision despite its large size necessitating complex real-time alignment measurements. Each chamber, as well as approximately 50 alignment reference bars in the endcap region, are equipped with CCD cameras, laser sources, and LED-illuminated masks which optically link chambers and bars in a three dimensional grid. This permits micron– level determination of chamber-to-chamber positions and chamber distortions. This information is used to correct drift tube positions and shape for muon track reconstruction. The endcap optical system produces about 8000 83 kB images during each 20 minute readout cycle. The optical data acquisition and image analysis are performed by a hardware/software system (LWDAQ) developed at Brandeis University. The system is segmented so that six processes running on several computers perform the optical readout and image analysis in parallel. We describe the architecture and implementation of the control system; monitoring of the optical readout processes; evaluation of the validity of images; display of results, validity parameters, and error conditions; and storage of analysis results and quality in an Oracle database. The distributed control architecture includes a Linux-based control and communication process and a PVSS SCADA system for the user interface, display functions, and database storage. Details of the architecture, communications, and performance will be presented.
Speaker: Craig Dowell (Univ. of Washington)
• 27
COOL Performance Tests and Optimization
The COOL software has been chosen by both Atlas and LHCb as the base of their conditions database infrastructure. The main focus of the COOL project in 2007 will be the deployment, test and validation of Oracle-based COOL database services at Tier0 and Tier1. In this context, COOL software development will concentrate on service-related issues, and in particular on the optimization of software performance for data insertion and retrieval from the database. This will involve several activities in parallel, such as improvements in the C++ client CPU consumption, improvements in the SQL query strategy for data insertion and retrieval from the database, and multi-client stress tests of different software releases and server setups. In this presentation, we will review the most important tests and optimizations which have been performed in this context.
Speaker: Marco Clemencic (European Organization for Nuclear Research (CERN))
• 28
Creating End-to-End Guaranteed Bandwidth Network Paths Across Multiple Domains with TeraPaths
Supporting reliable, predictable, and efficient global movement of data in high-energy physics distributed computing environments requires the capability to provide guaranteed bandwidth to selected data flows and schedule network usage appropriately. The DOE-funded TeraPaths project at Brookhaven National Laboratory (BNL), currently in its second year, is developing methods and tools that enable scientists to reserve network bandwidth for specific time windows and dedicate this bandwidth to important data transfers. The TeraPaths software does this through the creation of an end-to-end (or source computer host to destination computer host) virtual network path with guaranteed bandwidth for the duration of a data transfer. The path is set up through the direct configuration of network devices within end-site LANs and indirect configuration of WAN network devices through the automated invocation of WAN provider services. The software accommodates the end user with an easy to use and secure web interface, as well as an API, for submitting reservations. An ongoing development effort will provide access to TeraPaths services from within popular data transfer tools, and enlarge the current set of supported network devices. With the collaboration of ESnet and Internet 2 development teams, the TeraPaths-capable infrastructure, which currently extends from BNL to the University of Michigan, will be expanded to US ATLAS Tier 2 sites and beyond, with the goal of creating a production-quality data transfer environment for the benefit of the high-energy physics community.
Speakers: Dr Dantong Yu (Brookhaven National Laboratory) , Dr Dimitrios Katramatos (Brookhaven National Laboratory) , Dr Shawn McKee (University of Michigan)
• 29
Data Acquisition Backbone Core DABC
European FP6 program "HadronPhysics", JRA1 "FutureDAQ" contract number RII3-CT-2004-506078) For the new experiments at FAIR like CBM new concepts of data acquisition systems have to be developed like the distribution of self-triggered, time stamped data streams over high performance networks for event building. The DAQ backbone DABC is designed for FAIR detector tests, readout components test, data flow investigations, and DAQ controls. All kinds of data channels (front-end systems) are connected by plug-ins into functional components of DABC like data input, combiner, scheduler, event builder, analysis and storage. Several software packages (uDAPL, OpenFabric IB verbs, MPI2) have been used to measure the performance of InfiniBand on a cluster. One can achieve about 80% of the bidirectional nominal IB bandwidth (measured with up to 22 nodes). The XDAQ package developed for the CMS experiment has been chosen as infrastructure for DABC. The IB transport has been implemented in XDAQ. Measurements showed that the XDAQ transport causes too much overhead. Therefore a new faster (zero copy) transport layer is implemented. To be flexible in the selection of DAQ controls a DIM server has been implemented connecting the XDAQ InfoSpace to arbitrary DIM clients. For EPICS and LabView such clients are implemented.
Speaker: Dr Hans G. Essel (GSI)
• 30
Data Quality Monitoring for the CMS Electromagnetic Calorimeter
The electromagnetic calorimeter of the Compact Muon Solenoid experiment will play a central role in the achievement of the full physics performance of the detector at the LHC. The detector performance will be monitored using applications based on the CMS Data Quality Monitoring (DQM) framework and running on the High-Level Trigger Farm as well as on local DAQ systems. The monitorable quantities are organized into hierarchical structures based on the physics content. The information produced is delivered to client applications according to their subscription requests. The client applications process the received quantities, according to pre-defined analyses, thus making the results immediately available, and store the results in a database, and in the form of static web pages, for subsequent studies. We describe here the functionalities of the CMS ECAL DQM applications and report about their use in a real environment. In particular we detail the usage of the DQM during the data collection campaigns at the 2006 electron calibration test beams, at the cosmic muon calibration stand (2005-2007), at the CMS slice test (2006 Magnet Test and Cosmic Challenge), and during the installation and commissioning of the calorimeter in the CMS experimental area.
Speaker: Dr Giuseppe Della Ricca (Univ. of Trieste and INFN)
• 31
Data storage layout and architecture of the German T1
The grid era brings upon new and steeply rising demands in data storage. The GridKa project at Forschungszentrum Karlsruhe delivers its share of the computation and storage requirements of all LHC and 4 other HEP experiments. Access throughput from the worker nodes to the storage can be as high a 2 GB/s. At the same time a continuous throughput in the order of 300-400 MB/s into and out from GridKa must be guaranteed for several months without interruption. The scalable storage and networking concept is based on modular storage units that offer dCache, xrootd and NFS access to over 1000 clustered hosts. dCache directs over 300 pools with a total of 600 TB of disk storage. The tape connection via separate SAN is managed by Tivoli Storage Manager (TSM) using storage agents. The talk describes software and hardware components, their integration and interconnects, then focuses on the design criteria of the architecture. Plans for enhancements and directions based on current experiences, future expansion are also discussed.
Speaker: Dr Doris Ressmann (Forschungszentrum Karlsruhe)
• 32
Data Stream handling in the LHCb experiment
Events selected by LHCb's online event filtering farm will be assembled into raw data files of about 2 GBs. Under nominal conditions about 2 such files will be produced per minute. These files must be copied to tape storage and made available online to various calibration and monitoring tasks. The life cycle and state transitions of each files are managed by means of a dedicated data- base. Transfer of the files to permanent storage is using the Data Management System of DIRAC, slightly adapted for the purposes of the Online system. Once files are copied to CERNs mass-storage, replication to the LHCb Tier1 centres is initiated. This system has been designed for extreme robustness and reliability with redundance in almost every component and various failover mechanisms. The software, hardware and protocols designed for this purpose will be discussed in detail. Performance figures of data- challenges done in Spring 2007 will be shown.
Speaker: Dr Niko Neufeld (CERN)
• 33
Database architecture for the calibration of ATLAS Monitored Drift Tube Chambers
The calibration of the 375000 ATLAS Monitored Drift Tubes will be a highly challenging task: a dedicated set of data will be extracted from the second level trigger of the experiment and streamlined to three remote Tier-2 Calibration Centres. This presentation reviews the complex chain of databases envisaged to support the MDT Calibration and describes the actual status of the implementation and the tests that are being performed to ensure a smooth operation at the LHC start-up.
Speaker: Dr Manuela Cirilli (University of Michigan)
• 34
Dataharvester - a library for reading and writing hierarchic tuples'' from/to various file formats
A tool is presented that is capable of reading from and writing to several different file formats. Currently supported file formats are ROOT, HBook, HDF, XML, Sqlite3 and a few text file formats. A plugin mechanism decouples the file-format specific "backends" from the main library. All data are internally represented as "heterogenous hierarchic tuples"; no other data structure exists in the harvester. The tool is written in C++; a python interface exists, as well. It is a design goal of the tool to make writing tuples as simple as possible -- a feature that is intended make it a superb tool for debugging e.g. algorithmic code. To this end, the following features are implemented in the dataharvester: - Files are opened implicitly, not explicitly - the structure of the tuples is defined implicitly: defining and filling a tuple is one step only - the data types are defined implicitly, not explicitly. The dataharvester is fully autotooled. Debian packages exist; an rpm spec file is work in progress.
Speaker: Dr Wolfgang Waltenberger (Hephy Vienna)
• 35
Development of the Tier-1 Facility at Fermilab
CMS is preparing seven remote Tier-1 computing facilities to archive and serve experiment data. These centers represent the bulk of CMS's data serving capacity, a significant resource for reprocessing data, all of the simulation archiving capacity, and operational support for Tier-2 centers and analysis facilities. In this paper we present the progress on deploying the largest remote Tier-1 facility for CMS, located at Fermilab. We will present the development, procurement and operations experiences during the final year of preparation. We will discuss the results of scale tests and system design. We will outline the hardware selection and procurement and plans for the future to meet the needs of the experiment and the constraints of the physical facility. We will also discuss the successes and challenges associated with enabling a mass storage system to meet the various experimental needs at a significant increase in scale over what is currently achievable. Finally we will discuss the model to support US Tier-2 centers from the Tier-1 facility.
Speaker: Ian Fisk (Fermi National Accelerator Laboratory (FNAL))
• 36
Distributed database services in PHENIX - what it takes tosupport a Petabyte experiment.
After seven years of running and collecting 2 Petabytes of physics data, PHENIX experiment at the Relativistic Heavy Ion Collider (RHIC) has gained a lot of experience with database management systems ( DBMS ). Serving all of the experiment's operations - data taking, production and analysis - databases provide 24/7 access to calibrations and book-keeping information for hundreds of users at several computing centers worldwide and face the following challenges: - Simultaneous data taking, production and analysis result in hundreds of concurrent database connections and heavy server I/O load. - Online data production at remote sites requires a high degree of Master-Slave server synchronization. - Database size ( presently 100GB with half of data added in the last few months ) raises scalability concerns. - Long life of modern HENP experiments and fast development of database technologies make prediction of the best DBMS provider 5-10 years down the road difficult and require investments in design and support of good APIs. In this talk PHENIX solutions to the above problems will be presented and the trade-offs discussed.
Speaker: Irina Sourikova (BROOKHAVEN NATIONAL LABORATORY)
• 37
Efficient, Large Scale Data Transfers in Wide Area Network
The efficient use of high-speed networks to transfer large data sets is an essential component for many scientific applications including CERN’s LCG experiments. We present an efficient data transfer application, Fast Data Transfer (FDT), and a distributed agent system (LISA) able to monitor, configure, control and globally coordinate complex, large scale data transfers. FDT is an Application for Efficient Data Transfers which is capable of reading and writing at disk speed over wide area networks (with standard TCP). If it is used for memory to memory transfers between two computers it can saturate a 10Gb/s WAN link. Disk to disk transfers between several servers can saturate a 10Gb/s in both directions. It is written in Java, runs an all major platforms and it is easy to use. FDT is based on an asynchronous, flexible multithreaded architecture able to balance and optimize the access to disks and to control the flow of data in the network through multiple streams. It streams datasets continuously, using a managed pool of buffers through one several TCP sockets in parallel and is using independent threads to read and write on each physical device. LISA (The Localhost Information Service Agent) is lightweight dynamic services that provides complete system monitoring and is capable to dynamically configure the system or running applications. It provides the functionality to orchestrate and optimize distributed large data transfers at the Supercomputing 2006 Bandwidth Challenge between more than 200 systems around the world.
Speakers: Dr Iosif Legrand (CALTECH) , Ramiro Voicu (CALTECH)
• 38
Embedding Python- a new approach to specifying program options
Applications often need to have many parameters defined for execution. A few can be done with the command line, but this does not scale very well. I present a simple use of embedded Python that makes it easy to specify configuration data for applications, avoiding wiring in constants, or writing elaborate parsing difficult to justify for small, or one-off applications. But the capability extends far beyond that, in that the full power of Python is available for computation, consistency checking, importing default parameters.
Speaker: Prof. Toby Burnett (University of Washington)
• 39
End-to-End Network/Application Performance Troubleshooting Methodology
The computing models for LHC experiments are globally distributed and grid-based. In such a computing model, the experiments’ data must be reliably and efficiently transferred from CERN to Tier-1 regional centers, processed, and distributed to other centers around the world. Obstacles to good network performance arise from many causes and can be a major impediment to the success of this complex, multi-tiered data grid. Factors that affect overall network/application performance exist on the network end systems themselves (application software, operating system, hardware), in the local area networks that support the end systems, and within the wide area networks. Since the computer and network systems are globally distributed, it can be very difficult to locate and identify the factors that are hurting application performance. In this paper, we present an end-to-end network/application performance troubleshooting methodology developed and in use at Fermilab. The core of our approach is to narrow down the problem scope with a divide and conquer strategy. The overall complex problem is split into two distinct sub-problems: network end system diagnosis and tuning, and network path analysis. After satisfactorily evaluating, and if necessary resolving, each sub-problem, we conduct end-to-end performance analysis and diagnosis. The paper will discuss tools we use as part of the methodology. The long term objective of the effort is to enable end users to conduct much of the troubleshooting themselves, before (or instead of) calling upon network and end system “wizards,” who are always in short supply.
Speaker: Dr Wenji Wu (FERMILAB)
• 40
ETICS Meta-Data Software Editing - From Check Out To Commit Operations
People involved in modular projects need to improve the build software process, planning the correct execution order and detecting circular dependencies. The lack of suitable tools may cause delays in the development, deployment and maintenance of the software. Experience in such projects has shown that the arranged use of version control and build systems is not able to support the development of the software efficiently, due to the large number of errors that cause the breaking of the build process. In this paper, we describe a possible solution implemented in ETICS, an integrated infrastructure for the automated configuration, build and test of Grid and distributed software. ETICS has defined meta-data software abstractions, from which it is possible to download, build and test software projects, setting for instance dependencies, environment variables and properties. Furthermore, the meta-data information is managed by ETICS reflecting the version control system philosophy, thanks to the existence of a meta-data repository and the handling of a list of operations, such as check out and commit. Because of this, all the information related to a specific software are stored in the repository only when they are considered to be correct. By adopting this solution, we show a reduction of errors at build time. Moreover, by introducing this functionality, ETICS will be a version control system like for the management of the meta-data.
Speaker: Elisabetta Ronchieri (INFN CNAF)
• 41
Evaluating Disk Hardware for Small Deployments of PostgreSQL
The PostgreSQL database is a vital component of critical services at the RHIC/USATLAS Computing Facility such as the Quill subsystem of the Condor Project and both PNFS and SRM within dCache. Current deployments are relatively unsophisticated, utilizing default configurations on small-scale commodity hardware. However, a substantial increase in projected growth has exposed deficiencies in this model. Our goal, therefore, is to ensure the scalability and continued availability of our database servers while minimizing costs and administrative overhead. To attain this goal we tested database I/O throughput across a range of inexpensive server and local/external disk configurations in order to determine which was optimal for our environment. This evaluation considered processor type (AMD vs. Intel), disk family (SATA vs. SAS), RAID configuration, and the amount of system memory. Finally, while our evaluation was designed to solve specific problems, we believe that the results of our general tests can be applied to similar deployments of PostgreSQL elsewhere.
Speaker: Mr Alexander Withers (Brookhaven National Laboratory)
• 42
Evaluation of Goodness-of-Fit tests in physical use cases
The Statistical Toolkit provides an extensive collection of algorithms for the comparison of two data samples: in addition to the chisquared test, it includes all the tests based on the empirical distribution function documented in literature for binned and unbinned distributions. Some of these tests, like the Kolmogorov-Smirnov one, are widely used; others, like the Anderson-Darling or the Cramer-von Mises tests, have been previously used in more sophisticated physics data analyses; nevertheless, several of the tests in the Statistical Toolkit are largely unknown in high energy physics applications, and a few of them, like the weighted formulations of the Kolmogorov-Smirnov test, are available for the first time in an open-source software tool for statistical analysis. No systematic evaluation of the power of the goodness-of-fit tests has been documented so far in literature. The present work presents a comprehensive study of the power of all the goodness-of-fit tests in rich collection of the Statistical Toolkit in a variety of use cases specific to high energy physics experiments; it highlights the relative merits and deficiencies of the algorithms to deal with peculiar characteristics of the configurations under study. The results of this study provide guidance for the selection of the most appropriate algorithm in experimental analyses. The Statistical Toolkit user layer is interfaced to AIDA-compliant analysis tools and to ROOT.
Speaker: Dr Maria Grazia Pia (INFN Genova)
• 43
EVIO - A Lightweight Object-Oriented Event I/O Package
EVIO is a lightweight event I/O package consisting of an object-oriented layer on top of a pre-existing, highly efficient, C-based event I/O package. The latter, part of the JLab CODA package, has been in use in JLab high-speed DAQ systems for many years, but other underlying disk I/O packages could be substituted. The event format on disk, a packed tree-like hierarchy of banks, maps directly to XML, so notions such as stream and DOM parsing directly apply. The EVIO package transparently maps the packed binary representation on disk to/from an object hierarchy or DOM tree in memory. The in-memory tree can then be queried or modified using STL-like algorithms, function objects, etc. Utility programs can transform between binary and real XML (ASCII) format. EVIO will be used by the next generation of JLab online and offline software systems.
Speaker: Dr Elliott Wolin (Jefferson Lab)
• 44
Exercising CMS dataflows and workflows in computing challenges at the Spanish Tier-1 and Tier-2 sites
CMS undertakes periodic computing challenges of increasing scale and complexity to test its computing model and Grid computing systems. The computing challenges are aimed at establishing a working distributed computing system that implements the CMS computing model based on an underlying multi-flavour grid infrastructure. CMS dataflows and data processing workflows are exercised during a period of about a month targeting specific performance and scale goals. Performance values are measured, problems are identified and feedback into the design, integration and operation of the computing system is provided. The CMS computing architecture is based on a tier-organised structure of computing resources, based on a Tier-0 centre at CERN, 7 Tier-1 centres for organized mass data processing, and about 30 Tier-2 centres where user physics analysis is performed. The Tier-0 is in charge of storing the data coming from the detector onto mass storage, performs a prompt reconstruction of the data and distributes the data among the Tier-1 centres. The Tier-1 sites archive on mass storage its share of data, run data reprocessing, organized group physics analysis for data selection and distribute down the selected data to Tier-2's for user analysis. Tier-1 centres also have the responsibility of storing Monte Carlo data produced at the Tier-2 sites. The above mentioned workflows have been exercised during October 2006 at a scale of 25% of what is needed for operations in 2008, and at a 50% scale during July 2007. An overview of the data- and workflows conducted at the Spanish Tier-1 and Tier-2 sites during the CMS computing challenges since last CHEP conference is presented. The focus is on presenting achieved results, operational experience and lessons learnt during the challenges.
Speaker: Dr Jose Hernandez (CIEMAT)
• 45
Experience Running a Distributed Tier-2 for the ATLAS Experiment
The Spanish ATLAS Tier-2 is geographically distributed between three HEP institutes. They are IFAE (Barcelona) and IFIC (Valencia) and UAM (Madrid). Currently it has a computing power of about 400 kSI2k CPU, a disk storage capacity of 40 TB and a network bandwidth connecting the three sites and the nearest Tier-1 of 1 Gb/s. These resources will increase with time in parallel to those of all the ATLAS Tier-2. They will be about 875 kSI2k CPU and 387 TB disk capacity at the LHC startup in 2008. The main roles of the Tier-2 are to provide resources for Production of Simulated Events and Physics Distributed Data Analysis. Since 2002, it has been participating to the different Data Challenge exercises. Currently, it is achieving around 1.5% of the whole ATLAS production in the framework of the Computing System Commissioning exercise. On the other hand, a prototype of Distributed Data Analysis system is being tested, integrated, and deployed. Distributed Data Management is also arising as an important issue in the daily activities of the Tier-2. The distribution in three sites has shown to be useful due to an increasing service redundancy, a faster solution of problems and the share of computing expertise and know-how. Experience gained running the Distributed Tier-2 and the preparations to perform successfully at the LHC startup will be presented.
Speaker: Mr LUIS MARCH (Instituto de Fisica Corpuscular)
• 46
Experience with monitoring of Prague T2 Site
Each tier 2 site is monitored by various services from outside. The Prague T2 is monitored by SAM tests, GSTAT monitoring, RTM from RAL, regional nagios monitoring and experiment specific tools. Besides that we monitor our own site for hardware and software failures and middleware status. All these tools produce an output that must be regularly checked by site administrators. We will present our solution built on nagios that allows our administrators to check just one service (nagios) that encapsulates the results from all monitoring tools (outside and inside) and presents them in a single web page. The solution is based on simple plugin for every service. Nagios uses these plugins to check the results of monitoring tools. We use plugins developed at SRCE, RAL and Prague. We will also present automatic configuration scripts that allow us to generate nagios configuration from local database of servers and services.
Speaker: Tomas Kouba (Institute of Physics - Acad. of Sciences of the Czech Rep. (ASCR)
• 47
Farm Logbook, a tool to keep/trace all operations on a big computing farm
Every day operations on a big computer center farm like that of a Tier1 can be numerous. Opening or closing a host, changing batch system configuration, replacing a disk, reinstalling a host and so on, is just a short list of what can and will really happen. In these conditions remembering all that has been done could be really difficult. Typically a big farm is managed by a team so it can happen to forget which operations have been performed by some other colleague; in the worst case, one could be completely unaware about the changes. There is therefore a real need to keep trace as much as possible of all operations on the cluster in order to have the farm status under full control. With a tool like “Farm Logbook”, developed and deployed at the INFN Tier-1, you can easily track down information about events occurred in the past or ongoing. All information are obtained in an automatic way in order to avoid consistency problems due to someone missing updates. The technologies used to develop this tool, such as CVS, MySQL, PHP, AJAX and RSS, provide smart access to this information in order to grant anytime to farm administrators a realistic snapshot of the cluster. Potentially therefore this tool can improve farm management and troubleshooting.
Speaker: Mr Alessandro Italiano (INFN-CNAF)
• 48
FermiGrid - Experience and Future Plans
Fermilab supports a scientific program that includes experiments and scientists located across the globe. In order to better serve this community, Fermilab has placed its production computer resources in a Campus Grid infrastructure called 'FermiGrid'. The FermiGrid infrastructure allows the large experiments at Fermilab to have priority access to their own resources, enables sharing of these resources in an opportunistic fashion, and movement of work (jobs, data) between the Campus Grid and National Grids such as Open Science Grid and the WLCG. FermiGrid resources support multiple Virtual Organizations (VOs), including VOs from the Open Science Grid (OSG), EGEE and the Worldwide LHC Computing Grid Collaboration (WLCG). Fermilab also makes leading contributions to the Open Science Grid in the areas of accounting, batch computing, grid security, job management, resource selection, site infrastructure, storage management, and VO services. Through the FermiGrid interfaces, authenticated and authorized VOs and individuals may access our core grid services, the 10,000+ Fermilab resident CPUs, near-petabyte (including CMS) online disk pools and the multi-petabyte Fermilab Mass Storage System. These core grid services include a site wide Globus gatekeeper, VO management services for several VOs, Fermilab site authorization services, grid user mapping services, as well as job accounting and monitoring, resource selection and data movement services. Access to these services is via standard and well-supported grid interfaces. We will report on the user experience of using the FermiGrid campus infrastructure interfaced to a national cyberinfrastructure - the successes and the problems.
• 49
Fluka and Geant4 simulations using common geometry source and digitization algorithms
Based on the ATLAS TileCal 2002 test-beam setup example, we present here the technical, software aspects of a possible solution to the problem of using two differe! nt simulation engines, like Geant4 and Fluka, with ! the comm on geometry and digitization code. The specific use case we discuss here, which is probably the most common one, is when the Geant4 application is already implemented. Our goal then is to run the same simulation using the Fluka package by re-using the maximum number of the existing components. For simple setups, a tool (FLUGG) already exists that allows to use the Fluka engine while the navigation is performed with Geant4, starting from a description of the geometry in terms of Geant4 classes. In complex applications, however, the geometry is often built up at run time from the information stored in a database, and in these cases such a tool cannot be used directly; furthermore, it does not deal with sensitive detectors and digitization. We show how it is possible to overcome these two problems by building around FLUGG a set of tools ! for reading common Geometry Description Markup Language (GDML) files as well as for generating the output in the format allowing common processing algorithms.
Speaker: Manuel Gallas (CERN)
• 50
FPGA based Compute Notes for High Level Triggering in PANDA
PANDA is a new universal detector for antiproton physics at the HESR facility at FAIR/GSI. The PANDA data acquisition system has to handle interaction rates of the order of 10**7 /s and data rates of several 100 Gb /s. FPGA based compute nodes with multi-Gb/s bandwidth capability using the ATCA architecture are designed to handle tasks such as event building, feature extraction and high level trigger processing. Data connectivity is provided via optical links as well as multiple Gbit Ethernet ports. The boards will support trigger algorithms such us pattern recognition for RICH detectors, EM shower analysis, fast tracking algorithms and global event characterization. A high level hardware description language (Handel-C) will be used to implement the firmware.
Speaker: Prof. Wolfgang Kuehn (Univ. Giessen, II. Physikalisches Institut)
• 51
Glance Project: a database retrieval mechanism for the ATLAS detector
During the construction and commissioning phases of the ATLAS Collaboration, data related to the installation, testing and performance of the equipment are stored in distinctive databases. Each group acquires information and saves them in repositories placed in different servers, using diverse technologies. Both data modeling and terminology may vary among the storage areas. The development of retrieval systems for each data set would require too much effort and high maintenance cost. The goal of the Glance Project is to provide navigation mechanisms among the databases, which is independent of both the technology used to build them and their relationship. The browsing over the data sets results in hypertext tables and links on the Web. The user chooses the database and the system shows its structure. After selecting a partition of the repository, the system automatically creates a retrieval interface that allows the specification of search parameters. The interface can be customized for specific needs and saved in order to be easily accessed later. Glance corresponds to a single system that handles distinctive recovery mechanisms among diverse databases. Therefore, further knowledge about how data is organized and labeled is not required to perform queries. Maintenance costs are also minimized. This paper describes the Glance conception, its development and functionalities. The system usage is illustrated with some examples. Current status and future work are also discussed.
Speaker: Kathy Pommes (CERN)
• 52
GridKa Tier1 Site Management
GridKa is the German Tier1 centre in the Worldwide LHC Computing Grid (WLCG). It is part of the Institut für Wissenschaftliches Rechnen (IWR) at the Forschungszentrum Karlsruhe (FZK). It started in 2002 as the successor of the ”Regional Data and Computing Centre in Germany” (RDCCG) GridKa supports all four LHC experiments, ALICE, ATLAS, CMS and LHCb, four non-LHC high energy physics experiments as well as an astrophysics ex- periment and geclipse, a pro ject that concentrates on the development of tools for existing Grid infrastructures.. In this presentation we will give an overview of the site fabric management and fabric management tools. The usage of Rocks tool kit for worker node cluster installation is discussed. Most of the grid middleware services are deployed in Xen-based virtual machines on a variety of physical machines. We describe the integration of Yaim and Cfengine for installation and configu- ration of middleware services. We will also show how tools like Nagios and Cfengine can be combinedtoinitiaterecovery actions to ensure functionality of middleware services.
Speaker: Dr Sven Gabriel (Forschungszentrum Karlsruhe)
• 53
HEPTrails: An Analysis Workflow and Provenance Tracking Application
When doing an HEP analysis, physicists typically repeat the same operations over and over while applying minor variations. Doing the operations as well as remembering the changes done during each iteration can be a very tedious process. HEPTrails in an analysis application written in Python and built on top of the University of Utah's VisTrails system which provides workflow and full provenance tracking for scientific analysis. HEPTrails adds substantial extension to VisTrails in order to accommodate HEP analysis. These extensions include: a streaming workflow engine and with very fine grain modules, a lab notebook style presentation window capable of showing preliminary results, and local area distributed computing. Although HEPTrails is still in early development, it already shows great promise in aiding HEP physicists.
Speaker: Dr Christopher Jones (Cornell University)
• 54
High Performance Storage Tests At The INFN Pisa Computing Centre
We report about the tests performed in the INFN Pisa Computing Centre with some of the latest generation storage devices. Fibre Channel and NAS solutions have been tested in a realistic enviroment, both participating in Worldwide CMS's Service Challenges, and simulating analysis patterns with more than 500 jobs accessing concurrently]data files. Both usage pattern have evidentiated the ability to use today's storage links at 10 Gbit/s, allowing a steady transfer rate exceeding what a single Gbit/s interface cannot guarantee when the number of concurrent users increase over a few hundred nodes.
Speaker: Dr Enrico Mazzoni (INFN Pisa)
• 55
High-Performance Stream Computing for Particle Beam Transport Simulations
Understanding modern particle accelerators requires simulating charged particle transport through the machine elements. These simulations can be very time consuming due to the large number of particles and the need to consider many turns of a circular machine. Stream computing offers an attractive way to dramatically improve the performance of such simulations by calculating the simultaneous transport of many particles using dedicated hardware. Modern Graphics Processing Units (GPUs) are powerful and affordable stream computing devices. The results of simulations of particle transport through a FODO-cell transfer line, including an aperture model, using an NVidia GeForce 7900 GPU are compared to conventional transport codes. Accuracy and potential speed increases are compared and the prospects for future work in the area are discussed.
Speakers: Dr David Bailey (University of Manchester) , Dr Robert Appleby (University of Manchester)
• 56
HLRmon: a Role-based Grid Accounting Reporting Web Tool
In production quality Grid infrastructure accounting data play a key role on the possibility to spot out how the allocated resources have been used. The different types of Grid user have to be taken into account in order to provide different subsets of accounting data based on the specific role covered by a Grid user. Grid end users, VO (Virtual Organization) managers, site administrators and Grid operators need different ways to get resource usage statistics about jobs executed in a given time period at various different levels, depending on their specific Grid role. In the framework of the Italian production Grid, HLRmon aims at generating suitable reports for various categories of Grid users and has been designed to serve them within an unified layout. Thanks to its ability to authenticate web users through personal certificate and related authorization rights, it can a-priori restrict the selectable items range offered, so that proper information can only be provided to specifically enabled people. Information are gathered by HLRmon from the accounting database (HLR, Home Location Register), which stores complete accounting data in a per job basis. Depending on the kind of reports that are to be generated, it directly queries the HLR using an ad-hoc DGAS (Distributed Grid Accounting System) query tool (typically user's level detail info), or a local RDBMS table with daily aggregated information in a per day, site, VO basis, thus saving connection delay time and needless load on the HLR.
Speakers: Mr Enrico Fattibene (INFN-CNAF, Bologna, Italy) , Mr Federico Pescarmona (INFN-Torino, Italy) , Mr Giuseppe Misurelli (INFN-CNAF, Bologna, Italy) , Mr Stefano Dal Pra (INFN-Padova, Italy)
• 57
Implementation and Performance of the ATLAS Second Level Jet Trigger
ATLAS is one of the four major LHC experiments, designed to cover a wide range of physics topics. In order to cope with a rate of 40MHz and 25 interactions per bunch crossing, the ATLAS trigger system is divided in three different levels. The first one (LVL1, hardware based) identifies signatures in 2 microseconds that are confirmed by the the following trigger levels (software based). The Second Level Trigger (LVL2) only looks at a region of the space around the LVL1 signature (called Region of Interest or ROI), confirming/rejecting the event in about 10 ms, while the Event Filter (Third Level Trigger, EF) has potential full event access and larger processing times, of the order of 1 s. The jet selection starts at the LVL1 with dedicated processors that search for high ET hadronic energy depositions. At the LVL2, the jet signatures are verified with the execution of a dedicated, fast jet reconstruction algorithm. Given the fact that the main jet's background are jets,the energy calibration at the LVL2 is one of the major dificulties of this trigger, allowing to distinguish low/high energy jets. The algorithm for the calibration has been chosen to be fast and robust, with a good performance. The other major dificulty is the execution time of the algorithm,dominated by the data unpacking time due to the large sizes of the jet ROI. In order to reduce the execution time, three possible granularities have been proposed and are being evaluated: cell based (standard), energy sums calculated at each Fron-End Board (FEB) and the use of the LVL1 Trigger Towers. The FEB and Trigger Tower granularities are also being used/evaluated for the reconstruction of the missing ET triggers at the Event Filter, given the short times available to process the full event. In this presentation, the design and implementation of the jet trigger of ATLAS will be discussed in detail, emphasasing the major dificulties of each selection step. The performance of the jet algorithm, including timing, eficiencies and rates will also be shown, with detailed comparisons of the different unpacking modes.
Speaker: Dr Patricia Conde Muíño (LIP-Lisbon)
• 58
Implementing a Modular Framework in a Conditions Database Explorer for ATLAS
The ATLAS conditions databases will be used to manage information of quite diverse nature and level of complexity. The infrastructure in being built using the LCG COOL infrastructure and provides a powerful information sharing gateway upon many different systems. The nature of the stored information ranges from temporal series of simple values to very complex objects describing the configuration of systems like the TDAQ infrastructure including also associations to large objects managed outside of the database infrastructure. While an unified graphical user interface is crucial for browsing the different data, it must understand and display many different types of information in a flexible way suggesting the use of run time specific plugins. This extension mechanism was heavily used in the KTIDBEXPLORER application that defines and implements not only abstract interfaces to connections to databases and files but also supports extended ROOT and online configuration (OKS) objects in the database. The application is aware of the relations between database objects so it's relations (links) can be explored. The core application, built using QT, displays the hierarchical folder view and powerful table widgets with panels for selecting the database queries. An important example of this architecture is the oNline Objects extended Database browsEr (NODE), that is designed to access and display all data, including histograms and data tables, available in the ATLAS Monitoring Archive. To cope with the special nature of the monitoring objects, a plugin from the MDA framework to the Time managed science Instrument Databases (TIDB2) is used. The database browser is extended, in particular to include operations on histograms like display, overlap, comparisons as well as commenting and local storage.
Speaker: Antonio Amorim (Universidade de Lisboa (SIM and FCUL, Lisbon))
• 59
INFN Tier-1 status report
INFN CNAF is a multi experiment computing center acting as Tier-1 for LCG but also supporting other HEP and non HEP experiments and Virtual Organizations. The CNAF Tier-1 is one of the main Resource Centers of the Grid Infrastructure (WLCG/EGEE); the preferred access method to the center is through WLCG/EGEE and INFNGRID middleware and services. Critical issues to be addressed to meet the requirements of the LHC experiments are stability, robustness and scalability of the services provided: in particular technical infrastructural services like power supply and cooling, CPU, data handling systems and network infrastructure to cope with the amount of data foreseen in the LHC era, database and experiment-specific services, monitoring, authorization and accounting, and 24x7 support. In this poster we present the current status of the INFN Tier-1 infrastructure and services, an assessment of the experiences gained so far, and the expected evolution leading to the startup of LHC experiments.
Speaker: Luca dell'Agnello (INFN-CNAF)
• 60
Initial results from the MAGIC-II Datacenter at PIC
A new data center has been deployed for the MAGIC Gamma Ray Telescope, located in the Roque de los Muchachos observatory in the Canary Islands, Spain, at the Port d'Informació Científica in Barcelona. The MAGIC Datacenter at PIC recieves all the raw data produced by MAGIC, either via the network or tape cartridges, and provides archiving, rapid processing for quality control and calibration, reconstruction and generation of analysis datasets, and distribution of all levels of datasets to all institutes and users of the collaboration. In 2007, services will be expanded to include the submission of batch analysis by all members of the MAGIC collaboration. Additional services provided to MAGIC by PIC are a CVS server and Web-enabled databases for data catalogue, experimental conditions and calibration constants. Implementation of the Datacenter has required PIC to solve a number of challenges. One of them is the efficient mass storage handling for a large number of small files, which is accomplished by implementing Virtual Volumes based on ISO images stored on tape and mounted dynamically using automounter techniques. The other is coping with the limited WAN bandwidth available to the Roque de los Muchachos observatory and having to combine collection of raw data via network and tape cartridges. This paper summarizes the initial implementation of the MAGIC-II Datacenter at PIC, describes the main technologies deployed and gives the growth prospects for the next years.
Speaker: Prof. Manuel Delfino Reznicek (Port d'Informació Científica (PIC))
• 61
Integrated RT-AT-Nagios system at BNL USATLAS Tier1 computing center
Managing large number of heterogeneous grid servers with different service requirements posts great challenges. We describe a cost-effective integrated operation framework which manages hardware inventory, monitors services, raises alarms with different severity levels and tracks the facility response to them. The system is based on open source components: RT (Request Tracking) tracks user requests, AT (Asset Tracking) manages site inventory, while Nagios performs facility monitoring. We will discuss the integration of those components. The AT serves as central repository to store information about machines, services, groups of machines and services, their interdependencies and configuration. Problem reports sent to RT by users are reflected on asset history stored in AT database. Nagios system uses AT to obtain information about the components to be monitored. Detected problems are classified according to their severity, reported to experts and fed into RT system, where the progress towards their resolution is tracked. The paper will describe the AT data model, integration between AT and Nagios and interfacing the RT to other problem tracking systems. The described system provides a scalable solution to commission grid servers, automate the error-prone manual system configuration, and leverage the existing ticket system for problem tracking. It allows BNL to operate Tier1 facility 7X24, and meets service level agreements for each WLCG grid middleware component with different class of service requirements.
Speaker: Tomasz Wlodek (Brookhaven National Laboratory)
• 62
JetWeb: A library and server for Monte Carlo validation and data comparisons
Accurate modelling of high energy hadron interactions is essential for the precision analysis of data from the LHC. It is therefore imperative that the predictions of Monte Carlos used to model this physics are tested against existing and future measurements. These measurements cover a wide variety of reactions, experimental observables and kinematic regions. To make this process more reliable and easily verifiable, the CEDAR collaboration has developed a set of tools for tuning and validating models of such interactions, and for archiving the comparisons so that data from previous experiments can be easily compared to new models. JetWeb is a Java-based server which accesses SQL databases for data and model predictions, and acts as a gateway to LCG resources for CPU-intensive simulation jobs. The simulation jobs are run using the HZTool and Rivet applications, also developed within CEDAR, and the data for comparison is taken from HepData. The server already contains an array of useful data. The status, current application and view plans of the system are described.
Speaker: Jonathan Butterworth (University College London)
• 63
Large Scale Access Tests and Online Interfaces to ATLAS Conditions Databases
The ATLAS Trigger and Data Acquisition systems (TDAQ) to the Conditions databases has strong requirements on reliability and performance. Several applications were developed to support the integration of Condition database access with the online services in TDAQ like the interface to the Information Services and to the TDAQ configuration.. The DBStressor was developed to test and stress the access to the Conditions database using the LCG/COOL interface while operating in an integrated way as a TDAQ application. The performance of simultaneous Conditions database read accesses was studied in the context of the ATLAS High Level Trigger large computing farms. A large set of tests were performed involving up to 1000 computing nodes that accessed simultaneously the LCG central database server infrastructure at CERN. Most of the general figures of the results can be explained assuming a simple (model involving) described by a number of threads in the database servers that were providing data at near constant rate. The information storage requirements were the motivation for the ONline ASynchronous Interface form the Information Service (IS) with LCG/COOL databases. It avoids the backpressure from Online Database servers by managing a local cache. In parallel the OKS2COOL application was developed to store Configuration Databases into an Offline Database with history record.
Speaker: Antonio Amorim (Universidade de Lisboa (SIM and FCUL, Lisbon))
• 64
Lorentz Angle Calibration for the CMS Pixel Detector
The CMS Pixel Detector is hosted inside the large solenoid generating a magnetic field of 4 T. The electron-hole pairs produced by particles traversing the pixel sensors will thus experience the Lorentz force due to the combined presence of magnetic and electric field. This results in a systematic shift of the charge distribution. In order to achieve a high position resolution a correction for this shift, which can be up to 120$\mu$m, has to be applied. At start-up the Lorentz shift for a given bias voltage is well known from beam test studies. Due to irradiation the electric field in the sensors will change and thereby the Lorentz drift as well. Furthermore, since the irradiation will not be uniform across the detector, each sensor will be differently affected. Therefore, the effective Lorentz displacement will be regularly measured using data. We present a strategy to extract this drift by comparing the cluster shapes of pixel hits in fully reconstructed tracks. The procedure measures the Lorentz displacement as function of the sensor depth and is developed using the CMS simulation and reconstruction software.
Speaker: Vincenzo Chiochia (Universitat Zurich)
• 65
Low-level reconstruction software and regional unpacking for the CMS strip tracker
The CMS silicon strip tracker is unprecedented in terms of its size and complexity, providing a sensitive area of >200 m^2 and comprising 10M readout channels. Its data acquisition system is based around a custom analogue front-end ASIC, an analogue optical link system and an off-detector VME board that performs digitization, zero-suppression and data formatting. These data are forwarded to the CMS online computing farm, which performs reconstruction and provides the high-level trigger (HLT) using tools defined within the offline software framework. The strip tracker geometry and high-multiplicity events combine to create very large data volumes, which must be “unpacked” from a custom format into objects handled by the reconstruction chain. This must be done within stringent time quotas imposed by the available online computing resources. We review the issues and requirements for HLT, solutions for optimizing the low-level reconstruction chain, such as regional unpacking, and results based on simulation and the expected final detector configuration.
Speaker: Dr Robert Bainbridge (Imperial College London)
• 66
Management System of Event Processing and Data Files Based on XML Software Tools at Belle
The Belle experiment has been operational since 1999 and we have processed more than 700/fb of data so far. To cope with ever increasing data, complete automation of the event processing is one of the most critical issues. In addition, unified management in the processing job and the processed data files to be analyzed is very important especially to deal with ~400K data files amounting to Peta-scale data. To do this, we have implemented new application called “R&D Chain Management (RCM)”. In the RCM program, we define a “work- flow” as a set of check-points related to the event processing procedures such as job submission, job log verification, output file backup and so on. The work-flow is described in the XML and it is so flexible that we can cover all of actions required in the processing stream. The RCM application automatically surveys each check- point in the work-flow and it reports failure if any. Furthermore, simultaneously the RCM tools can store any necessary information at each check-point into the own database written in the XML. The RCM system provides the web-based interface between the XML database and users for an excellent visualization. Thus, one can quickly pick up relevant information from this system. The processed data files are also taken care by this RCM, and automatic action to take a backup, for instance, can be issued from the RCM tools. We will present our experience using the RCM software tool as well as performance in the Belle data management.
• 67
Metropolitan Area Network Support at Fermilab
Advances in wide area network service offerings, coupled with comparable developments in local area network technology have enabled many HEP sites to keep their offsite network bandwidth ahead of demand. For most sites, the more difficult and costly aspect of increasing wide area network capacity is the local loop, which connects the facility LAN to the wide area service provider(s). Fermilab has chosen to provide its own local loop access through leasing of dark fiber to a nearby network exchange point (StarLight), and procuring dense wave division multiplexing (DWDM) equipment to provide data channels across the fiber. Installing and managing such optical network infrastructure has broadened the Laboratory’s network support responsibilities to include operating network equipment that is located off-site, and is technically much different than classic LAN network equipment. Effectively, the Laboratory has assumed the role of a local service provider. This presentation will cover Fermilab’s experiences with deploying and supporting a Metropolitan Area Network (MAN) infrastructure, based on metro-range DWDM equipment, in order to meet its offsite networking needs. The benefits and drawbacks of providing and supporting such a service will be discussed. Issues of scalability, complexity, monitoring, and troubleshooting will also be discussed.
Speaker: Mr Philip DeMar (FERMILAB)
• 68
Monitoring a WLCG Tier-1 computing facility aiming at a reliable 24/7 service
Within the Worldwide LHC Computing Grid (WLCG), a Tier-1 centre like the German GridKa computing facility has to provide significant CPU and storage resources as well as several Grid services with a high level of quality. GridKa currently supports all four LHC Experiments, Alice, Atlas, CMS and LHCb as well as four non-LHC high energy physics experiments, and is about to significantly extend its services for other communities within the German Grid initiative D-Grid. In order to ensure the simultaneous usability of the resources by all VOs as well as the persistent import of data from CERN and the distribution of data to associated Tier-2 sites, a sophisticated monitoring model is essential. We present the GridKa monitoring concept which is based on the Ganglia and Nagios systems combined with additional tools to monitor Grid services and infrastructure. Due to the complex dependencies between a high number of monitored hosts and services, a clear and simple to use 'dashboard' showing a summarized view of the monitoring information is an essential tool. This 'dashboard' allows for a quick overview of the status and performance of services during the day and will be the first source of information for a deeper problem analysis if an automatic alarm notification is sent during nights and weekends.
Speaker: Dr Andreas Heiss (Forschungszentrum Karlsruhe)
• 69
The C++ reconstruction framework JANA has been written to support the next generation of Nuclear Physics experiments at Jefferson Lab in anticipation of the 12GeV upgrade. The JANA framework was designed to allow multi-threaded event processing with a minimal impact on developers of reconstruction software. As we enter the multi-core (and soon many-core) era, thread-enabled code will become essential to utilizing the full processor power available without invoking the logistical overhead of managing many individual processes. Event-based reconstruction lends itself naturally to mutli-threaded processing. Emphasis will be placed on the multi-threading features of the framework. Test results of the scaling of event processing rates with number of threads will be shown
Speaker: Dr David Lawrence (Jefferson Lab)
• 70
Nightly builds and software distribution in the LCG / AA / SPI project
The Software Process and Infrastructure project (SPI) of the LCG Applications Area (AA) is responsible for a set of services for software build, software packaging, software distribution, communication and quality assurance. Recently a new tool has been developed in SPI for the automatic configuration and build of the LCG AA software stack which is used for nightly builds. In this talk the design, features and performance of this nightly build system will be presented. Examples for configurations currently in use and their maintenance will be discussed as well as constraints and difficulties in the development of the system (e.g. multi platform, architectures, compilers, configuration tools, performance issues). The nightly build system will also be used to release the whole LHC software stack in one go. There will also be an outlook to future developments like distributed and parallel builds. The latter being important to speed up the delivery times for bug fixes of the LCG AA software stack to the user / experiments. This will become more important as we are nearing the startup of the LHC. The second part of the talk will describe policies of software packaging and software distribution in the LHC environment. As we are coming closer to the startup of the LHC it will be more and more important to be able to distribute the LHC software in an easy and transparent way to the users. The tools in the SPI project which aim to achieve this functionality and possibly can also be used by experiments for their software packaging and distribution will be presented.
Speaker: Dr Stefan Roiser (CERN)
• 71
Online Data Monitoring in the LHCb experiment
The High Level Trigger and Data Acquisition system selects about 2 kHz of events out of the 40 MHz of beam crossings. The selected events are sent to permanent storage for subsequent analysis. In order to ensure the quality of the collected data, indentify possible malfunctions of the detector and perform calibration and alignment checks, a small fraction of the accepted events is sent to a monitoring farm, which consists of a few tens of general purpose PCs. This contribution introduces the architecture of the data stream splitting mechanism from the storage system to the monitoring farm, where the raw data are analyzed by dedicated tasks. It describes the collaborating software components that are all based on the Gaudi event processing framework.
Speaker: Dr Markus Frank (CERN)
• 72
Optimization of dCache MSS tape efficiency through Virtual Volumes
Small files pose performance issues for Mass Storage Systems, particularly those using magnetic tape. The ViVo project reported at CHEP06 solved some of these problems by using Virtual Volumes based on ISO images containing the small files, and only storing and retrieving these images from the MSS. Retrieval was handled using Unix automounters, requiring deployment of ISO servers with a separately managed cache. We report developments which extend the use of ISO-based Virtual Volumes to the dCache storage management system, using Castor1 or ENSTORE as the tape back-end. By using the MSS interface already implemented and documented in dCache, we have been able to catalog files into dCache, pack them into ISO volumes, store the ISO images to tape and then transparently allow retrieval of the individual files by dCache. A simple catalog of files and corresponding ISO images allows us to fetch from tape the precise ISO image containing the requested file when it is requested by the dCache MSS interface. We also have developed a cache read-ahead technique that allows the injection of all files in an ISO image from MSS into dCache, triggered by an original single file request at the user level. This can provide a high tape efficiency if the files in each ISO image have been chosen to have a high probability of being requested within a short period of time. This also allows delegation back to dCache of most of the ISO-related (or Virtual Volume related) cache management, simplifying implementation and maintenance.
Speaker: Prof. Manuel Delfino Reznicek (Port d'Informació Científica (PIC))
• 73
Oracle RAC (Real Application Cluster) application scalability, experience with PVSS and methodology
Database applications increasingly demand higher performance. This is especially true in the context of the LHC accelerator, LHC experiments, and LHC Computing Grid projects at CERN. Oracle RAC (Real Application Cluster) is a cluster solution which allows a database to be served by several nodes, and is a technology that is being exploited successfully at CERN and at LCG Tier1 sites. Database applications often have initially low scalability, with a growing number of cluster nodes. This paper describes a methodology and innovative ideas developed in order to obtain almost linear scalability for some of the typical database workloads. This paper describes, amongst others, the work which has been performed on the PVSS “Oracle archiver” (the controls software used for LHC and its experiments), where the performance of the event archiving module has been increased from 1000 to 150000 event changes per second (x150). This has been achieved with several architectural changes (core program separation from the data manipulation, data loading techniques, and database schema). The result is also a near-linear scalability between the number of nodes in the cluster and the performance. Based on the experience gathered on many database projects, guidelines and tips are provided in order to help with the creation of scalable database applications, or to achieve scalability improvements for already designed database applications.
Speaker: Eric Grancher (CERN)
• 74
Parameterization of the LHCb magnetic field map
The LHCb warm magnet has been designed to provide an integrated field of 4 Tm for tracks coming from the primary vertex.To insure good momentum resolution of a few per mil, an accurate description of the magnetic field map is needed. This is achieved by combining the information from a TOSCA-based simulation and data from measurements. The paper presents the fit method applied to both the simulation and data to achieve the requirements. It also explains how the corresponding software tool is integrated in the LHCb Gaudi software and shows the relation with the environment in which it is used.
Speaker: Ms Geraldine Conti (EPFL)
• 75
Performance Measurement and Monitoring for HENP Applications
LHC experiments are entering in a phase where optimization in view of data taking as well as robustness' improvements are of major importance. Any reduction in event data size can bring very significant savings in the amount of hardware (disk and tape in particular) needed to process data. Another area of concern and potential major gains is reducing the memory size and I/O bandwidth requirements of processing nodes, especially with increasing usage of multi-core CPUs. LHC experiments are already collecting abundant performance information about event size, memory and CPU usage, I/O compression and bandwidth requirements. What is missing is a coherent set of tools to present this information in a tailored fashion to release coordinators, package managers and physics algorithm developers. This paper describes such a toolkit that we are developing in the context of ATLAS computing to harvest performance monitoring information from an extensible set of sources. The challenge is to map performance data with an immediate impact on hardware costs into entities which are relevant for the various users. For example the toolkit allows an ATLAS data model developer to evaluate the impact on resource usage throughout the entire software pipeline of their design decisions for an event data object. We present the data in a way that highlights potential areas of concerns, allowing experts to drill down to the level of detail they need (the size of a data member of a class, the CPU usage of a component method). A configurable monitoring system allows to set off alarms when a quantity or an histogram goes out of a specified range. This allows a release coordinator to monitor e.g. the global size of a data stream throughout a development cycle and have the developers correct a problem well before a release goes into production.
Speaker: Dr Sebastien Binet (LBNL)
• 76
Portable Monitoring for Castor2
We present our design, development and deployment of a portable monitoring system for the CERN Archival and Storage System (Castor) based on its existing internal database infrastructure and deployment architecture. This new monitoring architecture is seen as an important requirement for future development and support. Castor is now deployed at several sites which use different monitoring systems to the LHC Era Monitoring (Lemon) system used at CERN. This includes sites with significant computing resources in the United Kingdom and Italy. Providing a portable monitoring system is seen as desirable as this will reduce development overhead and provide a common framework for understanding the state of the systems and resolving operational tasks. We present an overview of the reasoning behind this project and its aims; a discussion on the various aspects of the system which have previously been monitored and how moving to this new system improves on this; and discuss development trade-offs and our future plans.
Speaker: Miguel Coelho Dos Santos (CERN)
• 77
Preparation for WLCG production from a Tier-1 viewpoint
The GRIDPP Tier-1 Centre at RAL is one of 10 Tier-1 centres worldwide preparing for the start of LHC data taking in late 2007. The RAL Tier-1 is expected to provide a reliable grid-based computing service running thousands of simultaneous batch jobs with access to a multi-petabyte CASTOR-managed disk storage pool and tape silo, and will support the ATLAS, CMS and LHCb experiments as well as many other experiments already taking or analysing data. The RAL Tier-1 is already well advanced towards readiness for LHC data-taking. We describe some of the reliability and performance issues encountered with various generations of storage hardware in use at RAL and how the problems were addressed. We describe the networking challenges for shipping late volumes of data into and out of the Tier-1 storage systems, and system to system within the Tier-1, and the changes made to accommodate the expected data volumes. We describe the scalability and reliability issues encountered with the grid-services and the various strategies used to minimise the impact of problems, including multiplying the number of service hosts, splitting services across a number of hosts, and upgrading services to more resilient hardware.
Speaker: Mr Martin Bly (STFC/RAL)
• 78
Real-time analysis of the operational state of the CMS strip tracker readout system
The CMS silicon strip tracker comprises a sensitive area of >200 m2 and 10M readout channels. Its data acquisition system is based around a custom analogue front-end ASIC, an analogue optical link system and an off-detector VME board that performs digitization, zero-suppression and data formatting. The data acquisition system uses the CMS online software framework, known as XDAQ, to configure, control and monitor the hardware components and steer the data acquisition. Recent developments have seen the integration of the CMS offline software framework, known as CMSSW, within the online data acquisition system. This provides many new features and services within the online environment, such as distributed analysis within CMSSW, access to geometry and conditions data, and a monitoring framework. We review how the monitoring frameworks available within both XDAQ and CMSSW will be used to assess the operational state of the hardware components of the strip tracker readout system during data-taking and provide real-time feedback to shifters in the CMS control room. We will report on the software components, the chosen architecture, the various monitoring streams available, and our experiences of commissioning and operating large-scale systems at the tracker integration facility.
Speaker: Dr Stefano Mersi (INFN & Università di Firenze)
• 79
Release Management - the EGEE project's approach
We describe an approach to maintaining a large integrated software distribution, the gLite middleware. We describe why we have moved away from the concept of regular releases of the entire distribution, favoring instead a multispeed approach where components can evolve at their own pace. An overview of our implementation of such a release process is given, explaining the full life cycle of updates as tracked from requirement capture through to deployment. In a broader context, the release strategies of comparable projects are assessed in an attempt to isolate some useful principles. We conclude with some thoughts on the future direction of releases for gLite.
Speaker: Dr Oliver Keeble (CERN)
• 80
Remote Management of nodes in the ATLAS Online Processing Farms
The ATLAS experiment will use of order three thousand nodes for the online processing farms. The administration of such a large cluster is a challenge especially due to high impact of any down time. The ability to quickly and remotely turn on/off machines, especially following a power cut, and the ability to monitor the hardware health whether the machine be on or off are some of the major issues which the ATLAS SysAdmin Team faced. To solve these problems ATLAS has decided wherever possible to use Intelligent Platform Management Interfaces (IPMI) for its nodes. This paper will present the mechanisms which were developed to allow the distribution of management and monitoring commands to the cluster machines in parallel. These commands were run simultaneously on the prototype farm and on the small scale final farm already purchased. The commands and their distribution take into account the specificities of the different IPMI versions and implementations, and the network topology of the ATLAS Online system. Results from timing measurements for the distribution of commands to many nodes will be shown. These measurements will cover the times for booting and for shutting down of the nodes and will be extrapolated to the final cluster size.
Speaker: Dr Marc Dobson (CERN)
• 81
Runtime memory and data monitoring using Reflex for memory usage evaluation and data quality control.
Runtime memory usage in experiments has grown enormously in recent years, especially in large experiments like Atlas. However, it is difficult to break down total memory usage as indicated by OS-level tools, to identify the precise users and abusers. Without a detailed knowledge of memory footprints, monitoring memory growth as an experiment evolves in order to control ballooning inflation, is ineffective. We present a process that make use of Reflex to traverse the contents of all objects in the event store. As each object is encountered, its constituents are analysed using Reflex based introspection, to determine their true memory footprint. Pointers to other constituent objects are recursively followed, and their full sizes are added to the total. Runtime configuration allows specific object types to be ignored. Simultaneously, the values of all data members can be printed, using either their fundamental types, or an optional user supplied method. Address maps are kept to ensure that multiple counting does not occur. By using these procedures, the analysis and reconstruction software can be monitored in a historical fashion, with statistics produced to show which objects and object containers grow or shrink in size, and likewise can be monitored for data quality consistency and release validation.
Speaker: Dr Charles Leggett (LAWRENCE BERKELEY NATIONAL LABORATORY)
• 82
Service Level Status - a new real-time status display for IT services
Nowadays, IT departments provide, and people use, computing services of an increasingly heterogeneous nature. There is thus a growing need for a status display that groups these different services and reports status and availability in a uniform way. The Service Level Status (SLS) system addresses these needs by providing a web-based display that dynamically shows availability, basic information and statistics about various IT services, as well as the dependencies between them. This paper first introduces the requirements SLS had to meet, and the main concepts behind it, like service availability and status, Key Performance Indicators (KPIs), sub/meta-services, and service dependencies. It then describes the SLS system architecture, and some interesting implementation details, such as the usage of XML Schemas. Since clear visualization of service availability and status is one of the main goals of SLS, emphasis is put on describing the intuitive web-based user interface.
Speaker: Mr Sebastian Lopienski (CERN)
• 83
StatPatternRecognition: A C++ Package for Multivariate Classification
SPR implements various tools for supervised learning such as boosting (3 flavors), bagging, random forest, neural networks, decision trees, bump hunter (PRIM), multi-class learner, logistic regression, linear and quadratic discriminant analysis, and others. Presented at CHEP 2006, SPR has been extended with several important features since then. The package has been stripped of CLHEP dependency, equipped with autotools and posted at Sourceforge for distribution under general public license: http://sourceforge.net/projects/statpatrec/ . It is now a standalone package with an optional dependency on Root for data input/output. Several new methods have been included in the package. SPR is now capable of boosting and bagging an arbitrary sequence of included classifiers allowing the user to explore a broad range of classifier combinations. This talk is meant to summarize recent updates to the package and review recent applications of the package to physics analysis. More info on the project is available from http://www.hep.caltech.edu/~narsky/spr.html .
Speaker: Dr Ilya Narsky (California Institute of Technology)
• 84
Survey of the main HEP software components (1970 --> 2010)
A poster (two A0 pages) shows the main software systems used in HEP in the period 1970 -> 2010 from their conception to their death. Graphics bands are used to indicate the relative importance of each system or tool in the following categories: -Machines and Operating systems -Storage systems and access libraries -Networking and communication software -Compiled languages -Code management systems -Data structures management and I/O systems. Data bases -Graphical User Interface and Graphics -Histograms, Math and Statistics -Scripting systems and Interpreters -Interactive Analysis systems -Detector Geometry and Simulation
Speaker: Dr Rene Brun (CERN)
• 85
Testing gLite for releases
We describe the methodology for testing gLite releases. Starting from the needs given by the EGEE software management process we illustrate our design choices for testing gLite. For certifying patches different test scenarios have to be considered: regular regression tests, stress tests and manual verification of bug fixes. Conflicts arise if these tests are all carried out at the same time on the same infrastructure. Thus virtualisation is used and its benefits are shown by several examples. Furthermore we sketch the architecture of our distributed testbed including lessons learnt from such a distributed test environment. Finally we describe the test framework we're using. We also give an overview of the tests that have been developed to test the different gLite services. Some statistics on the patches certified with this process are also presented. Apart from these more conventional testing activities we describe how we address testing maturing complex services for scalability and long term stability which require excessive testbed sizes.
Speaker: Mr Andreas Unterkircher (CERN)
• 86
The ATLAS Canada computing model consists of a Tier-1 computing centre located at the TRIUMF Laboratory in Vancouver, Canada, and two distributed Tier-2 computing centres: one in Eastern Canada and one in Western Canada. Each distributed Tier-2 computing centre is made up of a group of universities. To meet the network requirements of each institution, HEPnet Canada and CANARIE (Canada's National Research Network), in collaboration with other research networks, have constructed a series of lightpaths to connect the institutions involved. TRIUMF is connected to CERN via a 10G circuit, a portion of which will be dropped at TRIUMF's Tier-1 peer SARA/NIKEF in Amsterdam. Current Canadian ATLAS Tier-2 institutions are connected to TRIUMF via 1G lightpaths, and routing between Tier-2s occurs through TRIUMF. We discuss the architecture of the ATLAS Canada network, challenges of building the network, and future plans for the network.
Speaker: Mr Ian Gable (University of Victoria)
• 87
The ATLAS Software Installation System for LCG/EGEE
The huge amount of resources available in the Grids, and the necessity to have the most updated experiment software deployed in all the sites within a few hours, have spotted the need for automatic installation systems for the LHC experiments. In this paper we describe the ATLAS system for the experiment software installation in LCG/EGEE, based on the Lightweight Job Submission Framework for Installation (LJSFi). This system is able to automatically discover, check, install, test and tag the full set of resources made available in LCG/EGEE to the ATLAS Virtual Organization in a few hours, depending on the site availability. The installations or removals may be centrally triggered as well as requested by the end-users for each site. A fallback solution to the manual operations is also available, in case of problems. The installation data, status and job history are centrally kept in the installation DB and browseable via a web interface. The installation tasks are performed by one or more automatic agents. The ATLAS installation team is automatically notified in case of problems, in order to proceed with the manual operations. Each user may browse or request an installation activity in a site, directly by accessing the web pages, being identified by his personal certificate. This system has been successfully used by ATLAS since 2003 to deploy about 60 different software releases and has performed more than 75000 installation jobs so far. The LJSFi framework is currently being extended to the other ATLAS Grids (NorduGrid and OSG).
Speaker: Alessandro De Salvo (Istituto Nazionale di Fisica Nucleare Sezione di Roma 1)
• 88
The CMS Storage Manager and Data Flow in the High Level DAQ
With the turn-on of the LHC, the CMS DAQ system is expecting to log petabytes of experiment data in the coming years. The CMS Storage Manager system is a part of the high bandwidth event data handling pipeline of the CMS high level DAQ. It has two primary functions. Each Storage Manager instance collects data from the sub-farm, or DAQ slice of the Event Filter farm it has been assigned to, and logs it to disk. It also serves as an event and histogram server for calibration and monitoring processes. The Event Filter and Storage Manager cooperating systems use a special serialized form of the offline event data model in order to achieve the bandwidth specifications required by CMS. The online format is converted to the standard offline format in the tier zero before storage to the tape archive. This paper will detail the technical implementation and performance achievements of these cooperating systems during the recent data challenges.
Speaker: Ms Elizabeth Sexton-Kennedy (FNAL)
• 89
The CMS Tracker Control System
The Tracker Control System (TCS) is a distributed control software to operate 2000 power supplies for the silicon modules of the CMS Tracker and monitor its environmental sensors. TCS must thus be able to handle 10^4 power supply parameters, 10^3 environmental probes from the Programmable Logic Controllers of the Tracker Safety System (TSS), 10^5 parameters read via DAQ from the DCUs in all front end hybrids and from CCUs in all control groups. TCS is built on top of an industrial SCADA program (PVSS) extended with a framework developed at CERN (JCOP) and used by all LHC experiments. The logical partitioning of the detector is reflected in the hierarchical structure of the TCS, where commands move down to the individual hardware devices, while states are reported up to the root which is interfaced to the broader CMS control system. The system computes and continuously monitors the mean and maximum values of critical parameters and updates the percentage of currently operating hardware. Automatic procedures switch off selected parts of the detector using detailed granularity and avoiding widespread TSS intervention.
Speaker: Lorenzo Masetti (CERN)
• 90
The Design and Implementation of BES Grid Computing System
Beijing Electron Spectrometer (BESIII) experiment will produce 5 PB of data in next five years. Grid is used to solve this challenge. This paper introduces BES grid computing model and specific technologies, including automatic data replication, fine-grained job scheduling and so on.
Speaker: Prof. Gang Chen (IHEP, China)
• 91
The Distributed Parallel Multi-Platform ATLAS Release Nightly Build System
The ATLAS offline software comprises over 1000 software packages organized into 10 projects that are built on a variety of compiler and operating system combinations every night. File-level parallelism, package-level parallelism and multi-core build servers are used to perform simultaneous builds of 6 platforms that are merged into a single installation on AFS. This in turn is used to build distribution kits from which remote sites can create cloned installations. Since ATLAS typically has multiple release branches open simultaneously, corresponding to ongoing GRID productions, detector commissioning activities and ongoing software development, several instances of these distributed build clusters operate in parallel. We discuss the various tools that provide performance gains and the error detection and retry mechanisms that have been developed in order to counteract network and other instabilities that would otherwise degrade the robustness of the system.
Speaker: Obreshkov Emil (INRNE/CERN)
• 92
The Status of Grid for Belle Experiment
The Belle Experiment is an ongoing experiment with an asymmetric electron-positron collider at KEK and already has a few PB scales of data in total including hundreds TB DST (Data Summary Tape) and MC data. It’s too much difficult to export existing data to LCG (LHC Computing Grid) physically because of huge amount of data. We setup a SRB (Storage Resource Broker) server to access them by using SRB-DSI (SRB Data Storage Interface) which is an extension to the GridFTP server to interact with SRB. This approach isn’t benefit only for Belle LCG users but also for legacy users. The Belle VO (Virtual Organization) has been federated among 4 countries, 6 institutes, 9 sites, and the analysis software has been installed to every site of this VO. The results of testing functionality, measuring performance and future prospects based on this federation will be shown in this paper.
Speaker: Go Iwai (KEK/CRC)
• 93
The Swiss ATLAS Grid
Since 2005 the Swiss ATLAS Grid is in production. It comprises four clusters at one Tier 2 and two Tier 3 sites. About 800 heterogenous cores and 60 TB disk space are connected by a dark fibre network operated at 10 Giga bit per second. Three different operating systems are deployed. The Tier 2 cluster runs both LCG and NorduGrid middleware (ARC) while the Tier 3 clusters run only the latter. As local resource management systems both OpenPBS/Torque and Sun Grid Engine are used. In 2006 about 200 000 CPU hours of central ATLAS production were run on a smaller version of this infrastructure. Local users ran about 60 000 CPU hours for ATLAS physics studies. As end user tools ARC clients, ATLAS DQ2 software, ATLAS Ganga and a new graphical user interface are employed. With emphasis on the heterogenous solutions we report on the current and future infrastructure and usage of this part of the ATLAS grid.
Speaker: Mr Sigve Haug (LHEP University of Bern)
• 94
The Tier0 road to LHC Data Taking
CERN, as other sites, has been preparing computing services for the arrival of LHC data for some time---more than 11 years if everything started at the First LHC Computing Workshop, held in Padova in June 1996. With LHC data taking now just around the corner, this presentation takes a look back at preparations at CERN and considers some of the key choices made along the way. Which were prescient? Which will we live to regret? And, most importantly, are we ready?
Speaker: Dr Tony Cass (CERN)
• 95
The US LHCNet Network for HEP
In this paper we present the design, implementation and evolution of the mission-orientedUSLHCNet for HEP research. The design philosophy behind our network is to help meet the dataintensive computing challenges of the next generation of particle physics experiments with a comprehensive, network-focused approach. Instead of treating the network as a static, unchanging and unmanaged set of inter-computer links, we are developing and using it as a dynamic, configurable and closely monitored resource that is managed from end-to-end. We will present our work in the various areas of the project, recent changes in the infrastructure, including the addition of LCAS/VCAT/GFP capable SONET equipment, future plans, transport protocol research and grid application development. Our working methodology is a continous cycle of evaluating equipment and technologies (servers, networking equipment, new standards) and network application development in order to build a production network for research. Our goal is to construct a nextgeneration network that is able to meet the data processing, distribution, access and analysis needs of the particle physics community.
Speaker: Dan Nae (California Institute of Technology (CALTECH))
• 96
Towards a full implementation of a robust solution of a Domain Specific Visual Query Language DSVQL for HEP analysis
With the project PHEASANT a DSVQL was proposed for the purpose of providing a tools that could increase user's productivity while producing query code for data analysis. The previous project aimed at the proof concept and methodology feasability by introducing the concept of DSLs. We are now concetrated on implementation issues in order to deploy a final tool. The concept of domain specific languages has always been implicit in Software Engineering altough the development of such languages was never done in a systematic way. The main goal of having DSLs is to rise the level of abstraction, as the main idea is to provide the final user (stakeholder) tools to reason and model the solution by using concepts of the problem domain instead of having to reason with concepts of the problem domain ( meaning the implementation details like programming concepts and hardware restrictions). Once we have the model specifyed, we can use Model Driven Development and Software Product Lines techniques to deploy artifacts in a automatic way (meaning: software products, code, documentation etc).The SE community has been focusing its attention to methodologies and deploy tools for helping DSL developers in their effort to help productivity and effiency at several application domains such as HEP. These tools start to mature and it worths having a look in order to avoid "redoing the wheel". In this communication we will present the several technologies for DSLs meta-modeling studied in order to implement the DSVQL proposed by the PHEASANT project.
Speaker: Dr Patricia Conde Muíño (LIP-Lisbon)
• 97
Track based software package for measurement of the energy deposited in the calorimeters of the ATLAS detector
The measurement of the muon energy deposition in the calorimeters is an integral part of muon identification, track isolation and correction for catastrophic muon energy losses, which are the prerequisites to the ultimate goal of refitting the muon track using calorimeter information as well. To this end, an accurate energy loss measurement method in the calorimeters is developed which uses only Event Data Model tools and is used by the muon isolation tool in the official ATLAS software, in order to provide isolation related variables at the Event Summary Data level. The strategy of the energy deposition measurement of the track in the calorimeters is described. Inner Detector, or Muon Spectrometer tracks are extrapolated to each calorimeter compartment using existing tools, which take into account multiple scattering and bending due to the magnetic field. The energy deposited in each compartment is measured by summing-up cells, corrected for noise, inside a cone around the track of desired size. The results of the measured energy loss in the calorimeters with this method are validated with Monte Carlo single muon samples.
Speaker: Konstantinos Bachas (Aristotle University of Thessaloniki)
• 98
US CMS Tier-2 Computing
The CMS computing model relies heavily on the use of "Tier-2" computing centers. At LHC startup, the typical Tier-2 center will have 1 MSpecInt2K of CPU resources, 200 TB of disk for data storage, and a WAN connection of at least 1 Gbit/s. These centers will be the primary sites for the production of large-scale simulation samples and for the hosting of experiment data for user analysis -- an interesting mix of experiment-controlled and user-controlled tasks. As a result, there are a wide range of services that must be deployed and commissioned at these centers, which are responsible for tasks such as dataset transfer, management of datasets, hosting of jobs submitted through Grid interfaces, and several varieties of monitoring. We discuss the development of the seven CMS Tier-2 computing centers in the United States, with a focus on recent operational performance and preparations for the start of data-taking at the end of 2007.
Speaker: Kenneth Bloom (University of Nebraska-Lincoln)
• 99
Usage and Extension of Maven to build, release and distribute Java and Native Code Projects
Maven is a software project management and comprehension tool. Based on the concept of a project object model (POM), Maven can manage a project's build, reporting and documentation from a single XML file which declaratively specifies the project's properties. In short, Maven replaces Make or Ant, adds the handling of dependencies and generates documentation and a project website. Maven is an Open Source tool written in Java and mainly used for Java. It can easily be extended by writing common or project specific plugin modules. One of Maven's main features is the handling of your project's (transitive) binary dependencies. A simple declaration that your project needs a library and that library gets automatically downloaded from a nearby mirror of the Maven repository and installed for local usage. While downloading platform-independent Java libraries is relatively straightforward, Maven did not originally provide any support for the harder problem of dealing with platform specific native code dependencies. We extended Maven with the Native ARchive Plugin (NAR) to allow it to compile and link Native (C, C++ and Fortran) code and handle Native libraries as dependencies. In this talk we will describe the main features and benefits of Maven, how these benefits can be extended to native code using the NAR plugin, and describe how we use Maven as the project management tool for both Java and C++ components of the FreeHEP library.
Speaker: Mark Donszelmann (SLAC)
• 100
Usage of LSF for the batch farms at CERN
LSF 7, the latest version of Platform's batch workload management system, addresses many issues which limited the ability of LSF 6.1 to support large scale batch farms, such as the lxbatch service at CERN. In this paper we will present the status of the evaluation and deployment of LSF 7 at CERN, including issues concerning the integration of LSF 7 witht the gLite grid middleware suite and, in particular, the steps taken to endure an efficient reporting of the local batch system status and usage to the Grid Information System
Speaker: Dr Ulrich Schwickerath (CERN)
• 101
Use of Cfengine for deploying LCG/gLite middle-ware
Cfengine is a middle to high level policy language and autonomous agent for building expert systems to administrate and configure large computer clusters. It is ideal for large-scale cluster management and is highly portable across varying computer platforms, allowing the management of multiple architectures and node types within the same farm. As well as being a highly capable configuration manager, Cfengine can also augment a network based service manager by providing an alerting mechanism which can monitor the system for abnormal behaviour, eg. watching and maintaining the permissions on files and directories, to monitoring available file-system space. A number of sites within the GridPP collaboration and across the wider EGEE project are using Cfengine for the deployment and maintenance of their clusters and LCG/gLite middle-ware installations. This paper will collect together the experiences of these sites with both the initial setup and the ongoing maintenance of their Cfengine installations. This paper will also discuss some of the common implementation scenarios that may be encountered. This paper will also provide some solutions to common issues that the various sites have faced.
Speaker: Mr Colin Morey (Manchester University)
• 102
Use of Flow Data for Traffic Analysis and Network Performance Characterization
At Fermilab, there is a long history of utilizing network flow data collected from site routers for various analyses, including network performance characterization, anomalous traffic detection, investigation of computer security incidents, network traffic statistics and others. Fermilab’s flow analysis model is currently built as a distributed system that collects flow data from the site network border routers, as well as from internal core routers & aggregation switches. The flow data is complete, not sampled, with a daily volume of approximately 10GBytes. Despite the high volume of collected information, large scale analysis is conducted in near real-time to satisfy demands of the user community for timely availability of the analyzed data. In this paper, we present Fermilab's Netflow Collection and Analysis system, as well as tools developed to analyze the flow data. Tools presented will include traffic characterization and network performance estimation for the US-CMS Tier1 Center, verification of path symmetry for network traffic re-routed over alternate path circuits, and profiling of traffic patterns for individual systems to characterize their typical behavior and enable identification of anomalous behavior.
Speaker: Mr Andrey Bobyshev (FERMILAB)
• 103
User Centric Monitoring (UCM) information service for the next generation of Grid-enabled scientists
Nuclear and high-energy physicists routinely execute data processing and data analysis jobs on a Grid and need to be able to monitor their jobs execution at an arbitrary site at any time. Existing Grid monitoring tools provide abundant information about the whole system, but are geared towards production jobs and well suited for Grid administrators, while the information tailored towards an individual user is not readily available in a user-friendly and user-centric way. Such User Centric information includes monitoring information such as the status of the submitted job, queue position, time of the start/finish, percentage of being done, error messages, standard output, and reasons for failure. We proposed to develop a framework centered on Grid service technology that allows scientists to track and monitor their jobs easily from a user-centric view. The proposed framework aims to be flexible so that it can be applied by any Grid Virtual Organization (VO) with various ways of collecting the user-centric job monitoring information built into the framework. Furthermore, the framework provides a rich and reusable set of methods of presenting the information to the user from within a Web browser and other clients. In this presentation, we will give an architectural overview of the UCM service, show an example implementation in the RHIC/STAR experiment context and discuss limitations and future collaborative work.
Speaker: Dr David Alexander (Tech-X Corporation)
• 104
Using ROOT with Modern Programming Languages
ROOT is firmly based on C++ and makes use of many of its features – templates and multiple inheritance, in particular. Many modern languages like Java and C# and python are missing these features or have radically different implementations. These programming languages, however, have many advantages to offer scientists including improved programming paradigms, development environments, and full blown GUI development frameworks. Python is well served by the PyROOT project which gives full access to ROOT’s capabilities. The bindings between ROOT and Python are built on the fly by the PyROOT infrastructure from the CINT dictionaries. This poster reports on progress towards implementing a similar infrastructure for the C# and the .NET family of languages with an eventual goal towards helping with physics analysis.
Speaker: Prof. Gordon Watts (University of Washington)
• 105
Using the Grid for Large Scale and Nightly Testing of the ATLAS Trigger & Data Acquisition System
The ATLAS Trigger & Data Acquisition System has been designed to use more than 2000 CPUs. During the current development stage it is crucial to test the system on a number of CPUs of similar scale. A dedicated farm of this size is difficult to find, and can only be made available for short periods. On the other hand many large farms have become available recently as part of computing grids, leading to the idea of using them to test the TDAQ. However the task of adapting the TDAQ to run on the Grid is not trivial, as the TDAQ system requires full access to the computing resources it runs on and real-time interaction. Moreover the Grid virtualises the resources to present a common interface to the user. We will describe the implementation and first tests of a scheme that resolves these issues using a pilot job mechanism. The Tier2 cluster in Manchester was successfully used to run a full TDAQ system on 400 nodes using this implementation. Apart from the tests described above, this scheme also has great potential for other applications, like running Grid remote farms to perform detector calibration and monitoring in real-time, and automatic nightly testing of the TDAQ.
Speaker: Hegoi Garitaonandia Elejabarrieta (Instituto de Fisica de Altas Energias (IFAE))
• 106
VINCI : Virtual Intelligent Networks for Computing Infrastructures
The main objective of the VINCI project is to enable data intensive applications to efficiently use and coordinate shared, hybrid network resources, to improve the performance and throughput of global-scale grid systems, such as those used in high energy physics. VINCI uses a set of agent-based services implemented in the MonALISA framework to enable the efficient use of network resources, coordinated with computing and storage resources. VINCI is an integrated network service system that provides client authentication and authorization, discovery of services and the topology of connections, workflow scheduling, global optimization and monitoring. The distributed agent system provides dynamically, on demand, end to end optical (or layer two VLAN) connections in less than one second independent of the location and the number of optical switches involved. It monitors and supervises all the created connections and is able to automatically generate an alternative path in case of connectivity errors. The alternative path is set up rapidly enough to avoid a TCP timeout, and thus to allow the transfer to continue uninterrupted. Dedicated agents are used to monitor the client systems and to detect hardware and software configuration. They can perform end to end performance measurements and if necessary to configure the systems. We are developing agents able to interact with GMPLS and CIENA’s G.ASON network control plane protocols and to integrate this functionality into the network services provided by the VINCI framework.
Speaker: Prof. Harvey Newman (CALTECH)
• 107
Virtualization applications at the Brookhaven Computing Facility
The Brookhaven Computing Facility provides for the computing needs of the RHIC experiments, supports the U.S. Tier 1 center for the ATLAS experiment at the LHC and provides computing support for the LSST experiment. The multi-purpose mission of the facility requires a complex computing infrastructure to meet different requirements and can result in duplication of services with a large number of single-purpose servers for narrowly-defined applications. The facility is investigating potential applications of the open-source Xen virtualization package to allow the consolidation of services and servers. This presentation also discusses using Xen to virtualize the bulk of our Linux-based computing cluster. This is being integrated with Condor, the dCache-managed distributed storage system and the new multi-core CPU's to improve availability and increase effective usage of our facility resources by virtualizing a wide array of software support packages to meet the needs of various applications. Virtualization support (both hardware and software) can be an important element for efficient operations in a heterogeneous computing environment with increasing reliance on distributed computing models.
Speaker: Dr Tony Chan (BROOKHAVEN NATIONAL LAB)
• 108
Web System to support analysis for experimental equipment commissioning
During the ATLAS detector commissioning phase, installed readout electronics must pass performance standards tests. The resulting data must be analyzed to ensure correct operation. For the Tile Calorimeter, developers plug their code into a specific framework for physics data-processing,. Collaboration members, taking shifts on commissioning work, interpret the results, in thousands of readout channels, to identify potential problems that may need correction during commissioning. The Tile Commissioning Web System (TCWS) facilitates the repetitive data analysis and quality control by encapsulating all necessary steps to retrieve information, execute programs, access the outcomes, register statements, and verify the equipment status. TCWS integrates different applications, each presenting a particular view of the commissioning process. The TileComm Analysis application stores plots and analysis results, provides equipment-oriented visualization, collects information regarding equipment performance, and summarizes its status. The Timeline application provides equipment status history in a chronological way. The Web Interface for Shifters application supports monitoring tasks by managing test parameters, graphical views of the calorimeter performance, and information status of all equipment that was used in each test. Finally, equipment quality control data can be filled, stored, modified, and retrieved as hypertext forms through the ATLASMonitor application. These applications are also connected with other commissioning programs that allow an automatic gathering of the commissioning data. This paper describes in detail the programs that compose the TCWS and how they are integrated within the Tile Calorimeter commissioning. Current status and future work are also discussed
Speaker: Andrea Dotti (INFN)
• Plenary: Plenary 1 Carson Hall

Carson Hall

Convener: Mike Vetterli (Simon Fraser University / TRIUMF)
• 109
Opening Session
• 110
The LHC Machine and Experiments: Status and Prospects
The current status of the LHC machine and the experiments, especially the general-purpose experiments, will be given. Also discussed will be the preparations for the physics run in 2008. The prospects for physics, with an emphasis on what can be expected with an integrated luminosity of 1 fb-1, will be outlined.
Speaker: Tejinder Virdee (CERN/Imperial College)
• 111
WLCG - Where we are now and the Real Data Challenges ahead
The talk will review the progress so far in setting up the distributed computing services for LHC data handling and analysis and look at some of the challenges we face when the real data begins to flow.
Speaker: Les Robertson (CERN)
• 10:30 AM
Coffee Break
• Plenary: Plenary 2 Carson Hall

Carson Hall

Convener: Manuel Delfino (PIC)
• 112
LHC Expt Computing
Speaker: Dr Ian Fisk (FERMILAB)
• 113
Data Acquisition at the LHC experiments
The CERN Large Hadron Collider (LHC) is one of the most awesome science tool ever built. To fully exploit the potential of this great instrument, a huge design and development effort has been initiated in order to ensure that measurements can optimally flow out from the detectors in terms of quantity, selectivity, and integrity, be accessible for online monitoring and be recorded for analysis and long-term archive. This effort is now reaching the end of its initial development phase and evolving towards operation. We will give an overview of the resulting Trigger and Data Acquisition (DAQ) systems designed to harvest and store the precious stream of LHC data. We will in particular review some of the technology choices made to address the specific requirements of each experiment, covering both hardware and software aspects.
Speaker: Sylvain Chapeland (CERN)
• 114
Power, Density, Reliability & Performance: customer-driven evolution of cluster computing and storage, in high energy physics and other scientific fields.
Cluster systems now comprise 50% to 90% of the High Performance Computing (HPC) market. However, with computing and storage needs outpacing Moore's law, the traditional approach of scaling is giving rise to facility, administrative and performance issues. Details of industry trends and unmet customer requirements for cluster computing will be presented. Implications on systems and facility design will also be explored.
Speaker: Dr Eng Lim Goh (SGI)
• 12:30 PM
Lunch
• Collaborative tools: CT 1 Oak

Oak

Convener: Peter Clarke (National e-Science Centre, UK)
• 115
Collaborative Tools and the LHC: An Update
I report on major current activities in the domain of Collaborative Tools, focusing on development for the LHC collaborations and HEP, in general, including audio and video conferencing, web archiving, and more. This presentation addresses the follow-up to the LCG RTAG 12 Final Report (presented at CHEP 2006), including the formation of the RCTF (Remote Collaboration Task Force) to steer planning and development, the installation of prototype facilities at CERN, and current funding scenarios. I also summarize presentations and discussion made during the Shaping Collaboration 2006 conference held in Geneva in December 2006, and present proposals offered by the participants to specifically address major issues facing the LHC collaborations in the coming years.
Speaker: Dr Steven Goldfarb (University of Michigan)
• 116
EVO (Enabling Virtual Organizations), the Next Generation Grid-enable Collaborative
The EVO (Enabling Virtual Organizations) system is based on a new distributed and unique architecture, leveraging the 10+ years of unique experience of developing and operating the large distributed production based VRVS collaboration system. The primary objective being to provide to the High Energy and Nuclear Physics experiments a system/service that meet their unique requirements of usability, quality, scalability, reliability, and cost necessary for nationally and globally distributed research organizations. The EVO system, which will be officially released during March/April 2007 includes a better-integrated and more convenient user interface, a richer feature set including higher resolution video and instant messaging, greater adaptability to all platforms and operating systems, and higher overall operational efficiency and robustness. All of these aspects will be particularly important as we approach and then enter the startup period of the LHC because the community will require an unprecedented level of daily collaboration. There will be intense demand for long distance scheduled meetings, person-to-person communication, group-to-group discussions, broadcast meetings, workshops and continuous presence at important locations such as control rooms and experimental areas. The need to have the collaboration tools totally integrated in the physicists’ working environments will gain great importance. Beyond all these user-features, another key enhancement is the collaboration infrastructure network created by EVO, which covers the entire globe and which is fully redundant and resilient to failure. The EVO infrastructure automatically adapts to the prevailing network configuration and status, so as to ensure that the collaboration service runs without disruption. Because we are able to monitor the end-user’s node, we are able to inform the user of any potential or arising problems (e.g. excessive CPU load or packet loss) and, where possible, to fix the problems automatically and transparently on behalf of the user (e.g. by switching to another server node in the network, by reducing the number of video streams received, et cetera). The integration of the MonALISA architecture into this new EVO architecture was an important step in the evolution of the service towards a globally distributed dynamic system that is largely autonomous. The EVO system is intended to become the primary collaboration system used by the High Energy and Nuclear Physics community going forward.
Speaker: Mr Philippe Galvez (California Institute of Technology)
• 117
CERN Single Sign On solution
The need for Single Sign On has always been restricted by the lack of cross platform solutions: a single sign on working only on one platform or technology is nearly useless. The recent improvements in Web Services Federation (WS- Federation) standard enabling federation of identity, attribute, authentication and authorization information can now provide real extended Single Sign On solutions. CERN has investigated various options and now provides a Web SSO solution using some parts of WS-Federation technology. By using Shibboleth Service Provider module for Apache hosted web sites and Microsoft ADFS as identity provider linked to Active Directory user, users can now authenticate on any web application using a single authentication platform, providing identity, user information (building, phone...) as well as group membership enabling authorization possibilities. A typical scenario: a CERN user can now authenticate on a Linux/Apache website using Windows Integrated credentials, and his Active Directory group membership can be checked before allowing access to a specific web page.
Speaker: Mr Emmanuel Ormancey (CERN)
• 118
The Health-e-Child Project: A Grid enabled Platform for European Paediatrics
The Health-e-Child (HeC) project is an EC Framework Programme 6 Integrated Project that aims at developing an integrated healthcare platform for paediatrics. Through this platform biomedical informaticians will integrate heterogeneous data and perform epidemiological studies across Europe. The main objective of the project is to gain a comprehensive view of a child's health by ‘vertically’ integrating biomedical, information and knowledge that spans the entire spectrum from genetic to epidemiological. The resulting Grid enabled biomedical information platform will be supported by robust search, optimization and matching techniques for information collected in hospitals across Europe. In particular, paediatricians will be provided with decision support, knowledge discovery and disease modelling applications that will access data in hospitals in the UK, Italy and France, integrated via the Grid. For economy of scale, reusability, extensibility, and maintainability, Health-e-Child is being developed on top of an EGEE/gLite based infrastructure that provides all the common data and computation management services required by the applications. The emphasis of the Health-e-Child effort is on universality of information, person-centricity of information, universality of application, multiplicity and variety of biomedical analytics and person-centricity of interaction. Its corner stone is the integration of information across biomedical abstraction whereby layers of biomedical information (i.e., genetic, cell, tissue, organ, individual, and population layer) are integrated to provide a unified view of a person’s biomedical and clinical condition. This paper discusses the major issues and challenges in bio-medical data integration and how these will be resolved in the Health-e-Child system. It establishes the need for the HeC infrastructure and emphasises the importance of user requirements analysis when integrating highly heterogeneous medical information. HeC is presented as an example of how computer science originating from the high energy physics community can be adapted for use by biomedical informaticians to deliver tangible real-world benefits.
Speaker: Prof. Richard McClatchey (University of the West of England)
• 119
Benchmarks of medical dosimetry on the grid
Computational tools originating from high energy physics developments provide solutions to common problems in other disciplines: this study presents quantitative results concerning the application of HEP simulation and analysis tools, and of the grid technology, to dosimetry for oncological radiotherapy. The study concerned all the three major radiotherapy techniques: therapy with external beams from a medical linear accelerator (in particular, the modern technique or intensity modulated radiotherapy), brachytherapy with internal or superficial radioactive sources, and hadrontherapy with a proton beam. Geant4-based simulation applications developed for three realistic use case highlight the high precision dose calculation achievable; the simulation is complemented by AIDA-compliant analysis tools for the manipulation of dose distributions relevant to clinical usage. The application design exploits DIANE for transparent execution in sequential and parallel mode, either on a local farm or on the grid and GANGA as a grid user interface. Benchmarks for execution on the grid are presented; they highlight the capabilities and current limitations for exploiting the grid in real-life applications in all major branches of oncological radiotherapy. Computational tools for dosimetry based on HEP software systems couple high precision, speed suitable to clinical environments and low cost: therefore, they represent an alternative to commercial products of particular interest to developing countries for oncologial radiotherapy or radiation protection applications.
Speaker: Dr Maria Grazia Pia (INFN Genova)
• 120
CMS Offline Web Tools
We describe a relatively new effort within CMS to converge on a set of web based tools, using state of the art industry techniques, to engage with the CMS offline computing system. CMS collaborators require tools to monitor various components of the computing system and interact with the system itself. The current state of the various CMS web tools is described along side current planned developments. The CMS collaboration comprises of nearly 3000 people from all over the world. As well as its collaborators, its computing resources are spread all over globe and are accessed via the LHC grid to run analysis, large scale production and data transfer tasks. Due to the distributed nature of collaborators effective provision of collaborative tools is essential to maximise physics exploitation of the CMS experiment, especially when the size of the CMS data set is considered. CMS has chosen to provide such tools over the world wide web as a top level service, enabling all members of the collaboration to interact with the various offline computing components. Traditionally web interfaces have been added in HEP experiments as an afterthought. In the CMS offline we have decided to put web interfaces, and the development of a common CMS web framework, on an equal footing with the rest of the offline development. Tools exist within CMS to transfer and catalogue data (PhEDEx and DBS/DLS), run Monte Carlo production (ProdAgent) and submit analysis (CRAB). Effective human interfaces to these systems are required for users with different agendas and practical knowledge of the systems to effectively use the CMS computing system. The CMS web tools project aims to provide a consistent interface to all these tools.
Speaker: Giulio Eulisse (Northeastern University)
• Computer facilities, production grids and networking: CF 1 Carson Hall B

Carson Hall B

Convener: Kors Bos (NIKEF)
• 121
Lessons Learnt From WLCG Service Deployment
This talk summarises the main lessons learnt from deploying WLCG production services, with a focus on Reliability, Scalability, Accountability, which lead to both manageability and usability. Each topic is analysed in turn. Techniques for zero-user-visible downtime for the main service interventions are described, together with pathological cases that need special treatment. The requirements in terms of scalability are analysed, calling for as much robustness and automation in the service as possible. The different aspects of accountability - which covers measuring / tracking / logging / monitoring what is going on – and has gone on - is examined, with the goal of attaining a manageable service. Finally, a simple analogy is drawn with the Web in terms of usability - what do we need to achieve to cross the chasm from small-scale adoption to ubiquity?
Speaker: Dr Jamie Shiers (CERN)
• 122
Security Incidents management in a Grid environment
Today's production Grids connect large numbers of distributed hosts using high throughput networks and hence are valuable targets for attackers. In the same way users transparently access any Grid service independently of its location, an attacker may attempt to propagate an attack to different sites that are part of a Grid. In order to contain and resolve the incident, and since such an attack may rapidly span many different administrative domains, efficient sharing of appropriate information between sites is important. Appointing an incident coordinator to obtain, correlate, filter and redistribute relevant information needed by sites in their local investigations has proven to be very effective to rapidly process the massive amount of data involved. Improving the trust between the site security teams, which may have different security standards, is an important factor in obtaining more relevant and accurate information. However, wider distribution increases the risk of voluntarily or involuntarily leaks of sensitive information outside of the community. Such leaks are not only dangerous because they may expose sensitive information, but also because they may discourage other sites from sharing their findings in the future. As a result, an essential part of the incident response relies on processes implementing appropriate and timely management and control of the information flow. This document describes the model adopted by the EGEE infrastructure, as well as issues encountered.
Speaker: Dr Markus Schulz (CERN)
• 123
The Open Science Grid - Its Status and Implementation Architecture
The Open Science Grid (OSG) is receiving five years of funding across six program offices of the Department of Energy Office of Science and the National Science Foundation. OSG is responsible for operating a secure production-quality distributed infrastructure, a reference software stack including the Virtual Data Toolkit (VDT), extending the capabilities of the high throughput virtual facility, and supporting an expansion of the user base. OSG also educates existing and potential users. OSG Consortium members provide the computing and storage resources accessible from the distributed infrastructure and the user applications for its use. Over sixty DOE Lab and University facilities can now be accessed. Access to large storage resources is increasing. The infrastructure relies on ESNET and Internet2 production and advanced networks. The OSG implementation architecture presents the Virtual Organization (VO) - aka science/research community – as a capable middle tier between the diverse distributed resources and the end users. Implementation of this architecture focusses on: making each resource self-managed, secure, sharable, and accessible locally and remotely; providing secure common services, support and reference software to the communities to enable their effective use of the OSG Facility; and providing end-to-end and facility-wide tools, operational security, and user support. The OSG implementation architecture is cognizant of federated and intersecting infrastructures —spanning individually managed facilities, university department clusters, local area shared campus infrastructures, the large national grids, and community scoped distributed environments. We report on the status of OSG, its implementation architecture today, and plans for the future.
Speaker: Mrs Ruth Pordes (FERMILAB)
• 124
UK Grid Computing for High-Energy Physics
Over the last few years, UK research centres have provided significant computing resources for many high-energy physics collaborations under the guidance of the GridPP project. This paper reviews recent progress in the Grid deployment and operations area including findings from recent experiment and infrastructure service challenges. These results are discussed in the context of how GridPP is dealing with the important areas of networking, data storage and user education and support. Throughout, the paper offers feedback on observed successes and problems with the gLite middleware and the experiment specific software which is required to make the Grid usable. The paper moves on to examine current thinking on how ready GridPP sites are for LHC startup with an emphasis on some of the remaining challenges such as meeting strict WLCG availability targets and the scaling of resources at many sites to levels well beyond previous levels. The paper ends with a discussion of the decisions being taken to ensure a stable service and project through LHC startup until GridPP funding ends in 2011.
Speaker: Dr Jeremy Coles (RAL)
• 125
CDF offline computing'07: computing of a HEP experiment in a mature stage
CDFII detector at Fermilab is taking physics data since 2002. The architechture of the CDF computing system has substantially evolved during the years of the data taking and currently it reached stable configuration which will allow experiment to process and analyse the data until the end of Run II. We describe major architechtural components of the CDF offline computing - dedicated reconstruction and analysis farms, GRID-based Monte Carlo Production system, distributed databases, distributed hierarchical storage system, code development and distribution system. We present technical parameters of the CDF computing system and projected needs of the CDF computing for the next several years. We summarize the operational experience accumulated over the course of Run II and highlight the challenges the experiment had to overcome to reach the state where the CDF physicists are routinely using GRID in their daily analysis work.
Speaker: Dr Pavel Murat (Fermilab)
• 126
A Distriuted Tier-1 for WLCG
The Tier-1 facility operated by the Nordic DataGrid Facility (NDGF) differs significantly from other Tier-1s in several aspects: It is not located one or a few locations but instead distributed throughout the Nordic, it is not under the governance of a single organization but instead a "virtual" Tier-1 build out of resources under the control of a number of different national organizations. We present the technical implications of these aspects as well as the high-level design of this distributed Tier-1. The focus will be on computations, storage and monitoring. In order to look like a single site some services requires one single entry point. Most notably is the srm entry point to storage. This is archived using dCache, but with a number of additional features developed by NDGF and DESY/FermiLab. Computations are controlled by the NorduGrid ARC middleware, which has proven to be a reliable and easy to install a the sites and on top of that supports the distribution of the compute elements very well. Integration happens at the HEP experiment level, where one or more VO-boxes provide the interface between the experiments and the ARC middleware. Finally, diagnosing problems at sites where NDGF staff does not have administrative access to the machines requires a well developed set of monitoring services. This includes SAM tests especially tailored to the NDGF setup and use of other standard monitoring software packages.
Speaker: Mr Lars Fischer (Nordic Data Grid Facility)
• Distributed data analysis and information management: DD 1 Saanich

Saanich

Convener: Roger Jones (Lancaster University)
• 127
Ganga - a job management and optimising tool
Ganga, the job-management system (http://cern.ch/ganga), developed as an ATLAS- LHCb common project, offers a simple, efficient and consistent user experience in a variety of heterogeneous environments: from local clusters to global Grid systems. Ganga helps end-users to organise their analysis activities on the Grid by providing automatic persistency of the job's metadata. A user has full access to the job history including their configuration and input/output. It is however important that users can see a single environment for developing and testing algorithms locally and for running on large data samples on the Grid. The tool allows for some basic monitoring and a steadily increasing number of users of more than 300 users have been confirmed, both in HEP, as well as in non-HEP applications. The paper will introduce the Ganga philosophy, the Ganga architecture and current and future strategy. It will use the example of how an LHCb user performs his analysis using Ganga and will describe the experiences gathered so far with the tool in LHCb.
Speaker: Dr Andrew Maier (CERN)
• 128
ASAP is a system for enabling distributed analysis for the CMS Experiment
ASAP is a system for enabling distributed analysis for CMS physicists. It was created with the aim of simplifying the transition from a locally running application to one that is distributed across the Grid. The experience gained in operating the system for the past 2 years has been used to redevelop a more robust, performant and scalable version. ASAP consists of a client for job creation, control and monitoring and an optional server side component. Once jobs are delegated to the server it will submit, update, fetch or resubmit the job on behalf of the user. ASAP is able to make decisions on the success of the users job and will resubmit if either a grid or application failure is detected. An advanced mode allows running jobs to communicate directly with the server in order to request additional jobs and to set the status of the job directly. These features reduce the turnaround time experienced by the user and increase the likelihood of success.
Speaker: Dr Akram Khan (Brunel University)
• 129
The CERN Analysis Facility - A PROOF Cluster for Day-One Physics Analysis
ALICE (A Large Ion Collider Experiment) at the LHC plans to use a PROOF cluster at CERN (CAF - Cern Analysis Facility) for fast analysis. The system is especially aimed at the prototyping phase of analyses that need a high number of development iterations and thus desire a short response time. Typical examples are the tuning of cuts during the development of an analysis as well as calibration and alignment. Furthermore, the use of an interactive system with very fast response will allow ALICE to extract physics observables out of first data quickly. A test setup consisting of 40 machines exists for evaluation since May 2006. The PROOF system enables the distributed usage and xrootd the access to locally distributed files. An automatic staging system of files migrated to CASTOR and files available in the AliEn Grid has been developed. The talk will present the current setup as well as performance tests that have been performed. The integration of PROOF into ALICE's software framework (AliRoot) will be shown.
Speaker: Mr Jan Fiete Grosse Oetringhaus (CERN)
• 130
Distributed Data Analysis in LHCb
The LHCb distributed data analysis system consists of the Ganga job submission front-end and the DIRAC Workload and Data Management System. Ganga is jointly developed with ATLAS and allows LHCb users to submit jobs on several backends including: several batch systems, LCG and DIRAC. The DIRAC API provides a transparent and secure way for users to run jobs to the Grid and is the default mode of submission for the LHCb VO. This is exploited by Ganga to perform distributed user analysis for LHCb. This system provides LHCb with a consistent, efficient and simple user experience in a variety of heterogeneous environments and facilitates the incremental development of user analysis from local test jobs to the Worldwide LHC Computing Grid. With a steadily increasing number of users, the LHCb distributed analysis system has been tuned and enhanced over the past two years. This paper will describe the recent developments to support distributed data analysis for the LHCb experiment on WLCG.
Speaker: Dr Stuart Paterson (CERN)
• 131
Distributed Analysis using GANGA on the EGEE/LCG infrastructure
The distributed data analysis using Grid resources is one of the fundamental applications in high energy physics to be addressed and realized before the start of LHC data taking. The needs to manage the resources are very high. In every experiment up to a thousand physicist will be submitting analysis jobs into the Grid. Appropriate user interfaces and helper applications have to be made available to assure that all users can use the Grid without too much expertise in Grid technology. These tools enlarge the number of grid users from a few production administrators to potentially all participating physicists. The GANGA job management system (http://cern.ch/ganga), developed as a common project between the ATLAS and LHCb experiments provides and integrates these kind of tools. GANGA provides a simple and consistent way of preparing, organizing and executing analysis tasks within the experiment analysis framework, implemented through a plug-in system. It allows trivial switching between running test jobs on a local batch system and running large-scale analyzes on the Grid, hiding Grid technicalities. We will be reporting on the plug-ins and our experiences of distributed data analysis using GANGA within the ATLAS experiment and the EGEE/LCG infrastructure. The integration with the ATLAS data management system DQ2 into GANGA is a key functionality. In combination with the job splitting mechanism large amounts of jobs can be sent to the locations of data following the ATLAS computing model. GANGA supports tasks of user analysis with reconstructed data and small scale production of Monte Carlo data.
Speaker: Dr Johannes Elmsheuser (Ludwig-Maximilians-Universität München)
• 132
Integrating Xgrid technology into HENP distributed computing model
Modern Macintosh computers feature Xgrid, a distributed computing architecture built directly into Apple's OS X operating system. While the approach is radically different from those generally expected by the Unix based Grid infrastructures (Open Science Grid, TeraGrid, EGEE), opportunistic computing on Xgrid is nonetheless a tempting and novel way to assemble a computing cluster with a minimum of additional configuration. In fact, it requires only the default operating system and authentication to a central controller from each node. OS X also implements arbitrarily extensible metadata, allowing an instantly updated file catalog to be stored as part of the filesystem itself. The low barrier to entry allows an Xgrid cluster to grow quickly and organically. This paper and presentation will detail the steps that can be taken to make such a cluster a viable resource for HENP research computing. We will further show how to provide to users a unified job submission framework by integrating Xgrid through the STAR Unified Meta-Scheduler (SUMS), making tasks and jobs submission effortlessly at reach for those users already using the tool for traditional Grid or local cluster job submission. We will discuss additional steps that can be taken to make an Xgrid cluster a full partner in grid computing initiatives, focusing on Open Science Grid integration. MIT's Xgrid system currently supports the work of multiple research groups in the Laboratory for Nuclear Science, and has become an important tool for generating simulations and conducting data analyses at the Massachusetts Institute of Technology.
• Event processing: EP 1 Carson Hall A

Carson Hall A

Convener: Patricia McBride (Fermilab)
• 133
Simulation readiness for the first data at LHC
The ATLAS detector is entering the final phases of construction and commissioning in order to be ready to take data during the first LHC commissioning run, foreseen by the end of 2007. A good understanding of the experiment performance from the beginning is essential to efficiently debug the detector and assess its physics potential in view of the physics runs which are going to take place from 2008 on. The ATLAS Detector Simulation programs have been developed since the ATLAS inception and have been developed for easing the detector optimization and construction: further developments to the simulation suite have recently been introduced to cope with essential factors like misalignment, inefficiencies, imperfections but still maintaining a high level of efficiency and operability to serve the ongoing production exercises. Emphasis in this talk is put on recent developments and new features, on validation and production strategies as well as on performance figures, robustness and maintainability
Speaker: Prof. Adele Rimoldi (Pavia University & INFN)
• 134
Readiness of CMS Simulation towards LHC Startup
The CMS simulation based on the Geant4 toolkit and the CMS object-oriented framework has been in production for more than three years and has delivered a total of more than 200 M physics events for the CMS Data Challenges and Physics Technical Design Report studies. The simulation software has been successfully ported to the new CMS Event-Data-Model based software framework and is used in understanding the data taken by CMS in test beams as well as in the Magnet Test and Cosmic Challenge. In this paper, we present the experience from years in physics production, the migration process to the new architecture, efforts towards robustness and performance of the simulation chain and different operational scenarios in terms of hit simulation, event mixing and digitization.
Speaker: Sunanda Banerjee (Fermilab/TIFR)
• 135
The Pierre Auger Observatory Offline Software
The Pierre Auger Observatory aims to discover the nature and origins of the highest energy cosmic rays. The large number of physicists involved in the project and the diversity of simulation and reconstruction tasks pose a challenge for the offline analysis software, not unlike the challenges confronting software for very large high energy physics experiments. Previously we have reported on the design and implementation of a general purpose but relatively lightweight framework which allows collaborators to contribute algorithms and sequencing instructions to build up the variety of applications they require. In this report, we update the status of this work and describe some of the successes and difficulties encountered over the last few years of use. We explain the machinery used to manage user contributions, to organize the abundance of configuration files, to facilitate multi-format file handling, and to provide access to event and time-dependent detector information residing in various data sources. We also describe the testing procedures used to help maintain stability of the code in the face of a large number of contributions. Foundation classes will also be discussed, including a novel geometry package which allows manipulation of abstract geometrical objects independent of coordinate system choice.
Speaker: Thomas Paul (Northeastern University)
• 136
Intelligent Design
The International Linear Collider (ILC) promises to provide electron-positron collisions at unprecedented energy and luminosities. Designing the detectors to extract the physics from these events requires efficient tools to simulate the detector response and reconstruct the events. The detector response package, slic, is based on the Geant4 toolkit and adds a thin layer of C++ code. This allows the end user to fully describe the detector geometry and readout at runtime using a plain text file in an xml format which extends GDML. It also supports reading in simulated events in stdhep format, and writing out events in the ILC-standard LCIO format. We then describe org.lcsim, a Java toolkit for full event reconstruction and analysis. The components are fully modular and are available for tasks from digitization of tracking detector signals through to cluster finding, pattern recognition, track fitting, jetfinding, and analysis. The code can be run standalone, for batch or Grid computing, or from within JAS3, which then provides access to the WIRED event display and the AIDA-compliant analysis capabilities. We present the architecture as well as the implementation for several candidate detector designs, demonstrating both the flexibility and the power of the system.
Speaker: Norman Graf (SLAC)
• 137
The LDC Software Framework for the ILC detector
The International Linear Collider is the next large accelerator project in High Energy Physics. The Large Detector Concept (LDC) study is one of four international working groups that are developing a detector concept for the ILC. The LDC uses a modular C++ application framework (Marlin) that is based on the international data format LCIO. It allows the distributed development of reconstruction and analysis software. The framework is mainly used for optimizing the physics performance of the planned detector based on the Particle Flow paradigm. It recently has been adapted to also be applied in various detector prototype test beam studies within the EUDET project. In this talk we give an overview of the core framework where the focus will be on recent developments and improvements since it has been first presented at CHEP2006.
Speaker: Dr Frank Gaede (DESY IT)
• 138
Raw-data display and visual reconstruction validation in ALICE
ALICE Event Visualization Environment (AliEVE) is based on ROOT and its GUI, 2D & 3D graphics classes. A small application kernel provides for registration and management of visualization objects. CINT scripts are used as an extensible mechanism for data extraction, selection and processing as well as for steering of frequent event- related tasks. AliEVE is used for event visualization in offline and high-level trigger frameworks. The first emphasis of the talk is on visual representations of raw-data for different detector-types. Common infrastructure for thresholding and color-coding of signal/time information, placement of detector-modules in various 2D/3D layouts and for user-interaction with displayed data is presented. Methods for visualization of raw-data on different levels of detail are discussed as they are expected to play an important role during early detector operation with poorly understood detector calibration, occupancy and noise-levels. The second emphasis of the talk is put on tools developed for visual validation of reconstruction code on event-by- event basis. Since September 2006 ALICE applies a regular visual-scanning procedure to simulated proton-proton data to detect any shortcomings in cluster finding, tracking and primary & secondary vertex reconstruction. A high-level of interactivity is required to allow in-depth exploration of event-structure and navigation back to simulation records is supported for debugging purposes. Standard 2D projections and transformations are available for clusters, tracks and simplified detector geometry
• Grid middleware and tools: GM 1 Carson Hall C

Carson Hall C

Convener: Ian Bird (CERN)
• 139
Grid reliability
Thanks to the grid, users have access to computing resources distributed all over the world. The grid hides the complexity and the differences of its heterogeneous components. In order for this to work, it is vital that all the elements are setuped properly, and that they can interact with each other. It is also very important that errors are detected as soon as possible, and that the procedure to solve them is well established. Our goal is to improve the performance of the grid. In order to do this, we studied two of its main elements: the workload and the data management systems. We developed all tools needed to investigate the efficiency of the different centres. Furthermore, our tools can be used to categorize the most common error messages, and measure their time evolution. One common reason for job failures is site misconfiguration. Being able to detect such a misconfiguration as soon as possible helps in several ways: first of all, it minimizes the time that it takes to bring the site back to a normal state; moreover, debugging it is easier, since the problem happened in the recent past. This can be specially helpful for new centers, since the tools provide the material needed to get a better understanding of the grid's complexity. In this contribution we will describe all the tools that we have developed to monitor the grid efficiency. These tools are currently used by the four LHC experiments. We will also describe the results and benefits that the tools have provided.
Speaker: Pablo Saiz (CERN)
• 140
WLCG scale testing during CMS data challenges
The CMS computing model to process and analyze LHC collision data follows a data-location driven approach and is using the WLCG infrastructure to provide access to GRID resources. As a preparation for data taking beginning end of 2007, CMS tests its computing model during dedicated data challenges. Within the CMS computing model, user analysis plays an important role in the CMS computing strategy and poses a special challenge for the infrastructure with its random distributed access patterns. For this purpose, CMS developed the CMS Remote Analysis Builder (CRAB). CRAB handles all interactions with the WLCG infrastructure transparently for the user. During the 2006 challenge, CMS set its goal to test the infrastructure at a scale of 50,000 user jobs per day using CRAB. Both direct submissions by individual users and automated submissions by robots were used to achieve this goal. A report will be given about the outcome of the user analysis part of the challenge and observations made during these tests using both the EGEE and OSG parts of the WLCG will be presented. The test infrastructure will be described and improvements made during the challenge to reach the target scale will be discussed. In particular, the most prominent difference in the submission structure between both GRID middlewares will be discussed with regard to its impact on the challenge. EGEE uses a resource broker submission approach while OSG uses direct Condor-G submissions. For 2007, CMS plans to increase the scale of the tests by a factor of 2. A report on work done in 2007 in the context of preparation for the summer 2007 data challenge will be given and first results will be presented.
Speaker: Dr Oliver Gutsche (FERMILAB)
• 141
DZero Data-Intensive Computing on the Open Science Grid
High energy physics experiments periodically reprocess data, in order to take advantage of improved understanding of the detector and the data processing code. Between February and May 2007, the DZero experiment will reprocess a substantial fraction of its dataset. This consists of half a billion events, corresponding to more than 100 TB of data, organized in 300,000 files. The activity utilizes resources from sites around the world, including a dozen sites participating to the Open Science Grid consortium (OSG). About 1,500 jobs are run every day across the OSG, consuming and producing hundreds of Gigabytes of data. OSG computing and storage resources are coordinated by the SAM-Grid system. This system organizes job access to a complex topology of data queues and job scheduling to clusters, using a SAM-Grid to OSG job forwarding infrastructure. For the first time in the lifetime of the experiment, a data intensive production activity is managed on a general purpose grid, such as OSG. This paper describes the implications of using OSG, where all resources are granted following an opportunistic model, the challenges of operating a data intensive activity over such large computing infrastructure, and the lesson learned throughout the few months of the project.
Speaker: Dr Amber Boehnlein (FERMI NATIONAL ACCELERATOR LABORATORY)
• 142
PanDA: Distributed production and distributed analysis system for ATLAS
A new distributed software system was developed in the fall of 2005 for the ATLAS experiment at the LHC. This system, called PanDA, provides an integrated service architecture with late binding of jobs, maximal automation through layered services, tight binding with ATLAS distributed data management (DDM) system, advanced error discovery and recovery procedures, and other features. In this talk, we will describe the PanDa software system. Special emphasis will be placed on the evolution of PanDA based on one and half year of real experience in carrying out CSC data production for ATLAS. The architecture of Panda is well suited for the computing needs of the ATLAS experiment, which is expected to be one of the first HEP experiments to operate at the petabyte scale.
Speaker: Tadashi Maeno (Brookhaven National Laboratory)
• 143
AliEn2: the ALICE grid Environment
Starting from the end of this year, the ALICE detector will collect data at a rate that, after two years, will reach 4PB per year. To process such a large quantity of data, ALICE has developed over the last seven years a distributed computing environment, called AliEn, integrated in the WLCG environment. The ALICE environment presents several original solutions, which have shown their viability in a number of large exercises of increasing complexity called data challenges. This talk will describe the architecture of the ALICE distributed computing environment, focusing on the challenges to be faced and on the technical solutions chosen to meet them. The job submission system and the data management infrastructure will be described, as well as the user interface. The current status of AliEn will be illustrated, as well as the performance of the system during the data challenges. The presentation will describe also the development roadmap of the system.
Speaker: Dr Pablo Saiz (CERN)
• 144
Storage Resource Manager version 2.2: design, implementation, and testing experience
Storage Services are crucial components of the Worldwide LHC Computing Grid (WLCG) infrastructure spanning more than 200 sites and serving computing and storage resources to the High Energy Physics LHC communities. Up to tens of Petabytes of data are collected every year by the 4 LHC experiments at CERN. To process these large data volumes it is important to establish a protocol and a very efficient interface to the various storage solutions adopted by the WLCG sites. In this work we report on the experience acquired during the definition of the Storage Resource Manager v2.2 protocol. In particular, we focus on the study performed to enhance the interface and make it suitable for use by the WLCG communities. At the moment 5 different storage solutions implement the SRM 2.2 interface: BeStMan (LBNL), CASTOR (CERN, RAL, and INFN), dCache (DESY and FNAL), DPM (CERN), and StoRM (INFN and ICTP). After a detailed inside review of the protocol, various test suites have been written identifying the most effective set of tests: the S2 test suite from CERN and the SRM-Tester test suite from LBNL. Such test suites have helped verifying the consistency and coherence of the proposed protocol and validating existing implementations. We conclude our work describing the results achieved.
Speaker: Dr Flavia Donno (CERN)
• Software components, tools and databases: SC 1 Lecture

Lecture

Convener: Dirk Duellman (CERN)
• 145
ATLAS Analysis Model
As we near the collection of the first data from the Large Hadron Collider, the ATLAS collaboration is preparing the software and computing infrastructure to allow quick analysis of the first data and support of the long-term steady-state ATLAS physics program. As part of this effort considerable attention has been payed to the "Analysis Model", a vision of the interplay of the software design, computing constraints, and various physics requirements. An important input to this activity has been the experience of Tevatron and B-Factory experiments, one topic which was explored discussed in the ATLAS October 2006 Analysis Model workshop. Recently, much of the Analysis Model has focused on ensuring the ATLAS software framework supports the required manipulations of event data; the event data design and content is consistent with foreseen calibration and physics analysis tasks; the event data is optimized in size, access speed, and is accessible both inside and outside the software framework; and that the analysis software may be developed collaboratively.
Speaker: Dr Amir Farbin (European Organization for Nuclear Research (CERN))
• 146
CABS3 - CLEO Analysis by Script @ Belle
We developed the original CABS language more than 10 years ago. The main objective of the language was to describe a decay of a particle as simply as possible in the context of usual HEP data analysis. A decay mode, for example, can be defined as follows: define Cand Dzerobar kpi 2 { K+ identified pi- identified } hist 1d inv_mass 0 80 1.5 2.3 all momentum'' cut inv_mass .ge. 1.83 .and. inv_mass .le. 1.90 ... end-define where Dzerobar kpi together defines a particle made of K+ and pi-. At the execution time of the translated C++ code, the list of Dzerobar candidate combinations are generated from the list of K+ and pi- candidate lists. The reason we invented the scripting language was the following. Although C++ allows us to define new class and operators for them and let us write native C++ code like "Dzerobar = Kplus + piminus" to make combinations, we thought there was no way to extend it to the cases with more than two particles; Dzerobar = Kplus + piminus + piplus + piminus to make combinations out of four particles as the language only defines the binary operator. After more than ten years, the author learned that it is indeed possible to define set of classes in C++ and write a single line of code to automatically generate the list of candidate combinations from generic cases with N(>0) particles. The same particle can appear more than once at any order. This noble idea comes from the delayed evaluation of a purely functional language, Haskell. CABS3 is purely templated so that it incorporates any List class derived from <vector>.
Speaker: Prof. Nobuhiko Katayama (High Energy Accelerator Research Organization)
• 147
Physics Analysis Tools for CMS experiment at LHC
By end of 2007 the CMS experiment will start running and Petabytes of data will be produced every year. To make analysis of this huge amount of data possible the CMS Physics Tools package builds the highest layer of the CMS experiment software. A core part of this package is the Candidate Model providing a coherent interface to different types of data. Standard tasks like combinatorial analyses, generic cuts, MC truth matching and fitting are supported. Advanced template techniques enable the user to add missing features easily. We explain the underlying model, certain details of implementation and present some use cases showing how the tools are currently used in generator and full simulation studies as preparation for analysis of real data.
Speaker: Luca Lista (INFN Napoli)
• 148
HepData: restructuring the archive for LHC data
The Durham HepData database has for many years provided an up-to-date archive of published numerical data from HEP experiments worldwide. In anticipation of the abundance of new data expected from the LHC, the database is undergoing a complete metamorphosis to add new features and improve the scope for use of the database by external applications. The core of the HepData restructuring is the use of a relational database server in place of the legacy hierarchical system, and the use of a Java object model to abstract the database operations into object relationships. Additionally, an XML dialect, HepML, has been developed to describe data records: this provides a rich description of HEP datasets for use by the database migration system and by experiments wishing to submit their own data records. Standard Java persistency systems are used both for the object-relational and object-XML mappings.A new user front end is being developed, using Java Web application technology. This will provide easy user access to HepData's records and flexible output formats, including data comparisons. Furthermore, methods are in development to allow experimental collaborations to input and maintain their own data in a secure way.The re-development of HepData is part of the CEDAR project, which also involves the JetWeb and Rivet event generator tuning systems. HepData's role as a reference source for generator tunings is pivotal to the success of JetWeb and Rivet. In this paper we describe the current status of the development of the new HepData database.
Speaker: James William Monk (Department of Physics and Astronomy - University College London)
• 149
ETICS: the international software engineering service for the grid
The ETICS system is a distributed software configuration, build and test system designed to fulfill the needs to improve the quality, reliability and interoperability of distributed software in general and grid software in particular. The ETICS project is a consortium of five partners (CERN, INFN, Engineering Ingegneria Informatica, 4D Soft and the University of Wisconsin- Madison). The ETICS service consists of a build and test job execution system based on the NMI/Metronome software and an integrated set of web services and software engineering tools to design, maintain and control build and validation scenarios. The ETICS system allows taking into account complex dependencies among applications and middleware components and provides a rich environment to perform static and dynamic analysis of the software and execute deployment, system and interoperability tests. This presentation gives an overview of the system architecture and functionality set and then describes how the EGEE, DILIGENT and OMII-Europe projects are using the software engineering services to build, validate and distribute their software. Finally a number of significant use and test cases will be described to show how ETICS can be used in particular to perform interoperability tests of grid middleware using the grid itself.
Speaker: Dr Alberto Di Meglio (CERN)
• 150
The CMS Dataset Bookkeeping Service
The CMS Dataset Bookkeeping Service (DBS) has been developed to catalog all CMS event data from Monte Carlo and Detector sources. It includes the ability to identify MC or trigger source, track data provenance, construct datasets for analysis, and discover interesting data. CMS requires processing and analysis activities at various service levels and the system provides support for localized processing or private analysis, as well as global access for CMS users at large. Catalog entries can be moved among the various service levels with a simple set of migration tools, thus forming a loose federation of databases. DBS is available to CMS users via a Python API, Command Line, and a Discovery web page interfaces. The system is built as a multi-tier web application with Java servlets running under Tomcat, with connecting via JDBC to Oracle or MySQL database backends. Clients connect to the service through HTTP or HTTPs with authentication provided by GRID certificates and authorization through VOMS. DBS is an integral part of the overall CMS Data Management and Workflow Management systems. The system has been in operation since March 2007, an overview of the schema, functionality, deployment details, operational statistics and experience will be presented.
Speaker: Dr Lee Lueking (FERMILAB)
• 4:00 PM
Coffee Break
• Computer facilities, production grids and networking: CF 2 Carson Hall B

Carson Hall B

Convener: Kors Bos (NIKEF)
• 151
PetaCache: Data Access Unleashed
The PetaCache project started at SLAC in 2004 with support from DOE Computer Science and the SLAC HEP program. PetaCache focuses on using cost-effective solid state storage for the hottest data under analysis. We chart the evolution of metrics such as accesses per second per dollar for different storage technologies and deduce the near inevitability of a massive use of solid- state storage in the near future. We report on the latency and access-rate performance of a DRAM-based prototype constructed in 2005 using commodity hardware and a Flash-based prototype constructed in 2007 using purpose-built hardware. We describe the use of xrootd to cluster individual servers in a highly scalable and fault tolerant approach, and present tests of scalability. A study of access to ATLAS AOD data is reported as a first step in understanding the software issues that will be encountered in achieving unfettered access to objects within HEP events. Finally we examine the cost-benefit outlook for the use of solid-state storage in HEP experiments.
Speaker: Dr Richard Mount (SLAC)
• 152
CASTOR2: design and development of a scalable architecture for a hierarchical storage system at CERN
In this paper we present the architecture design of the CERN Advanced Storage system (CASTOR) and its new disk cache management layer (CASTOR2). Mass storage systems at CERN have evolved over time to meet growing requirements, both in terms of scalability and fault resiliency. CASTOR2 has been designed as a Grid-capable storage resource sharing facility, with a database-centric architecture, to keep the whole status of the system, and stateless daemons. We present an overview of the software architecture upon which CASTOR2 daemons are built, and the UML based software process that is in place to speed up and automate code development. We also demonstrate how external policies may be plugged into the framework to ease the operation of a CASTOR2 system, which is now being used in production at CERN as well as at a number of Tier1 sites since more than one year.
Speaker: Dr Giuseppe Lo Presti (CERN/INFN)
• 153
Experiences with gStore, a scalable Mass Storage System with Tape Backend
GSI in Darmstadt (Germany) is a center for heavy ion research and hosts an Alice Tier2 center. For the future FAIR experiments at GSI, CBM and Panda, the planned data rates will reach those of the current LHC experiments at Cern. Since more than ten years gStore, the GSI Mass Storage System, is successfully in operation. It is a hierarchical storage system with a unique name space. Its core consists of several tape libraries from different vendors and currently ~20 data mover nodes connected within a SAN network. The gStore clients transfer data via fast socket connections from/to the disk cache of the data movers (~40TB currently). Each tape is accessible from any data mover, fully transparent to the users. The tapes and libraries are managed by commercial software (IBM Tivoli Storage Manager TSM), whereas the disk cache management and the TSM and user interfaces are provided by GSI software. For Alice users all gStore data are worldwide accessible via Alice grid software, and in a test environment the Alice Tier2 xrootd system has been integrated successfully with gStore. For 2007 it is planned to provide ~200TB via xrootd backed with gStore. Our experiences show that it's possible to develop, maintain and operate successfully a large scale mass storage system with mainly 2 FTEs. As gStore is completely hardware independent and fully scalable in data capacity and I/O bandwidth, we are optimistic to fulfill also the dramatically increased mass storage requirements of the FAIR experiments in 2014, which will be several orders of magnitude higher than today.
Speaker: Dr Horst Goeringer (GSI)
• 154
Advances in Integrated Storage,Transfer and Network Management
UltraLight is a collaboration of experimental physicists and network engineers whose purpose is to provide the network advances required to enable and facilitate petabyte-scale analysis of globally distributed data. Existing Grid-based infrastructures provide massive computing and storage resources, but are currently limited by their treatment of the network as an external, passive, and largely unmanaged resource. This paper will give an overview of the recent advances made within the UltraLight collaboration over the last 3 years within the different work areas of the project which include: the UltraLight testbed, transportation layer (FAST TCP and MAX net), transfer applications (FDT), network aware command and control systems (VINCI), network centric storage clouds (LStore), and physics applications (data streaming and distributed analysis). Several of the technologies developed within the UltraLight project are currently being deployed and field tested to support efficient transfer of data and prepare for LHC startup. The core of the tools rely on globally distributed publish and subscribe infrastructures and end-to-end monitoring, to prevent single points of failure, increase robustness and improve scalability.
Speaker: Paul Avery (University of Florida)
• 155
Distributed Cluster dynamic storage: A comparison of dcache, xrootd and slashgrid storage systems running on batch nodes.
The HEP department of the University of Manchester has purchased a 1000 nodes cluster. The cluster is dedicated to run EGEE and LCG software and is currently supporting 12 active VOs. Each node is equipped with 2x250 GB disks for a total amount of 500 GB and there is no tape storage behind nor raid arrays are used. Three different storage solutions are currently being deployed to exploit this space: dcache, xrootd, slashgrid (HTTP based). In this paper we will present a comparison of their ease of use and their performance from different perspectives: System management perspective (ease of installation and maintainance); user perspective (type of functionality and reliability); and from the performance point of view, with random and streamed access of files. The comparisons have been done with different conditions of load on the worker nodes, and with different file sizes. Test executables, user analysis jobs accessing data from real HEP experiments and file transfers clients have been used in these tests.
Speaker: Ms Alessandra Forti (University of Manchester)
• Distributed data analysis and information management: DD 2 Lecture

Lecture

Convener: Roger Jones (Lancaster University)
• 156
Particle accelerators produce huge amounts of information in every experiment and such quantity cannot be stored easily in a personal computer. For that reason, most of the analysis is done using remote storage servers (this will be particularly true when the Large Hadron Collider starts its operation in 2007). Seeing how the bandwidth has increased in the last few years, the biggest problem of this approach at the moment is latency, which hurts considerably the performance of the analysis process. Fortunately, particle events are independent of each other, which allows us to transfer the information that must be processed in the future while analyzing the data at hand. The independent nature also allows us to transfer many events instead of the single one needed at a given time. Such ideas are implemented in the data analysis framework ROOT, and its file servers (rootd, xrootd and a http plugin). Among the techniques used, we have pre-reads, pre-fetching, parallel streams and atomic readings for multiple requests. All these strategies present an enormous advantage and will facilitate processing remote files at almost the same speed as local ones, as long as the bandwidth does not present any limitations.
Speaker: Leandro Franco (CERN)
• 157
Scaling CMS data transfer system for LHC start-up
The CMS experiment will need to sustain uninterrupted high reliability, high throughput and very diverse data transfer activities as the LHC operations start. PhEDEx, the CMS data transfer system, will be responsible for the full range of the transfer needs of the experiment. Covering the entire spectrum is a demanding task: from the critical high-throughput transfers between CERN and the Tier-1 centres, to high-scale production transfers among the Tier-1 and Tier-2 centres, to managing the 24/7 transfers among all the 170 institutions in CMS and to providing straightforward access to handful of files to individual physicists. In order to produce the system with confirmed capability to meet the objectives, the PhEDEx data transfer system has undergone rigourous development and numerous demanding scale tests. We have sustained production transfers exceeding 1 PB/month for several months and have demonstrated core system capacity several orders of magnitude above expected LHC levels. We describe the level of scalability reached, and how we got there, with focus on the main insights into developing a robust, lock-free and scalable distributed database application, the validation stress test methods we have used, and the development and testing tools we found practically useful.
Speaker: Lassi Tuura (Northeastern University)
• 158
Data management in BaBar
The BaBar high energy experiment has been running for many years now, and has resulted in a data set of over a petabyte in size, containing over two million files. The management of this set of data has to support the requirements of further data production along with a physics community that has vastly different needs. To support these needs the BaBar bookkeeping system was developed, and within this datasets are defined for data access and use. Datasets are defined in such a way to keep data separate for the hundreds of concurrent analyses, produced from many production cycles, and to keep similar data together for any specific use. In the development of this system, data has been modeled as a flow of information, that constantly changes. This system has been in use now for many years, and has been very successful in meeting these disparate needs. The methods for defining and managing datasets which will undergo constant changes will be discussed. The needs of production also require the distribution of data to computing centers, and the control of production with datasets will be mentioned. With the needs of a constantly changing dataset, the ability to analyze data from a known state, and then add to the analysis changes in the dataset at a future time will also be presented.
Speaker: Dr Douglas Smith (Stanford Linear Accelerator Center)
• 159
DIRAC: Data Production Management
The LHCb Computing Model describes the dataflow model for all stages in the processing of real and simulated events and defines the role of LHCb associated Tier1 and Tier2 computing centres. The WLCG ‘dressed rehearsal’ exercise aims to allow LHC experiments to deploy the full chain of their Computing Models, making use of all underlying WLCG services and resources, in preparation for real data taking. During this exercise simulated RAW physics data, matching the properties of eventual real data, will be uploaded from the LHCb Online storage system to Grid enabled storage. This data will then be replicated to LHCb Tier1s and subsequently processed (reconstructed and stripped). The product of this processing is user analysis data that are distributed to all LHCb Tier1 sites. DIRAC, LHCb’s Workload and Data Management System, supports the implementation of the Computing Model in a data driven, real time and coordinated fashion. In this paper the LHCb Computing Model will be reviewed and the DIRAC components providing the needed functionality to support the Computing Model will be detailed. The experience gained during WLCG's 'dressed rehearsal' exercise will be given along with an evaluation of the preparedness for real data taking.
Speaker: Andrew Cameron Smith (CERN)
• Event processing: EP 2 Carson Hall A

Carson Hall A

Convener: Stephen Gowdy (SLAC)
• 160
The Geant4 Virtual Monte Carlo
The Virtual Monte Carlo (VMC) provides the abstract interface into the Monte Carlo transport codes: Geant3, Geant4 and Fluka. The user VMC based application, independent from the specific Monte Carlo codes, can be then run with all three simulation programs. The VMC has been developed by the ALICE Offline Project and since then it draw attention in more experimental frameworks. Since its first release in 2002, the implementation of the VMC for Geant4 (Geant4 VMC) is in continuous maintenance and development, mostly driven by the requirements from new, non ALICE, users. In this presentation we will give an overview and the present status of this interface. We will report on new features, such as support for user defined Geant4 classes (the physics list, detector construction class) or support for Root TGeo geometry definition and G4Root navigation. We will also discuss the aspects specific to Geant4 and make a reflection about the strong and weak points of the VMC approach.
Speaker: Dr Ivana Hrivnacova (IPN)
• 161
A GEANT4 based simulation for proton therapy
The GEANT4 Monte Carlo code provides many powerful functions for conducting particle transport simulations with great reliability and flexibility. GEANT4 has been extending the application fields for not only the high energy physics but also medical physics. Using the reliable simulation for the radiation therapy, it will become possible to validate treatment planning and select the most effective one. For the use of a simulation in the clinical application, the simulation has to reproduce the dose distributions in three-dimensions with the best accuracy for ensuring the patient safety. As a generalized simulator, the GEANT4 based simulation framework has been developed and used for the verification of the simulated dose distribution to the measurements. Three types of irradiation systems for proton therapy were successfully implemented on the top of this framework; those are the gantry treatment nozzle at the Hyogo Ion Beam Medical Center (HIBMC), the gantry treatment nozzle at the National Cancer Center (NCC), and the eye treatment facility of UC San Francisco at the Crocker Nuclear Laboratory cyclotron, UC Davis (CNL). The validation of the simulation was performed for the proton ranges in important materials at beam line and the size of radiation field, respectively. Then dose distributions in simulation were verified with measurements for Bragg peak and spread out Bragg peak, respectively. We will report a belief description of the developed simulation, and the comparisons of simulated dose distributions with measurements as well as the validation of the beam irradiation system.
Speaker: Dr Tsukasa Aso (Toyama National College of Maritime Technology, JST CREST)
• 162
The performance of the Geant4 Standard EM package for LHC and other applications
Current status of the Standard EM package of the Geant4 toolkit is described. The precision of simulation results is discussed with the focus on LHC experiments. The comparisons of the simulation with the experimental data are shown.
Speaker: Prof. Vladimir Ivantchenko (CERN, ESA)
• 163
Final results of the precision validation of Geant4 models in the pre-equilibrium and nuclear de-excitation phase
A project is in progress for a systematic, quantitative validation of Geant4 physics models against experimental data. Due to the complexity of Geant4 physics, the validation of Geant4 hadronic models proceeds according to a bottom-up approach (i.e. from the lower energy range up to higher energies): this approach, which is different from the one adopted in the LCG Simulation Validation Project, allows establishing the accuracy of individual Geant4 models specific to a given energy range on top of already validated models pertinent to a lower energy. Results are presented concerning the lower energy hadronic interaction phases: the nuclear de-excitation and pre-equilibrium (up to 100 MeV). All relevant Geant4 electromagnetic and hadronic physics models, and pre-packaged physics configurations distributed by the Geant4 Collaboration (PhysicsLists) have been included in the validation test. The hadronic models for inelastic scattering involve Nuclear De-excitation in two variants (default and GEM), Precompound (with or without Fermi break-up), Bertini and Binary Cascade, and parameterised models. Elastic scattering includes parameterised models and the newly developed Bertini Elastic model. Various prepackaged PhysicsLists are also subject to the same validation process. The validation is performed against experimental data measured with 2% accuracy. The quantitative comparison of simulated and experimental data distributions exploits a rigorous goodness-of-fit statistical analysis. The final results from high statistics production on the grid are presented: they compare both the relative accuracy and the execution performance of all the options considered. These results provide guidance to users about the choice of Geant4 electromagnetic and hadronic physics models.
Speaker: Dr Maria Grazia Pia (INFN Genova)
• 164
Fast shower simulation in ATLAS Calorimeter
The simulation of the ATLAS detector is largely dominated by the showering of electromagnetic particles in the heavy parts of the detector, especially the electromagnetic barrel and endcap calorimeters. Two procedures have been developed to accelerate the processing time of EM particles in these regions: (1) a fast shower parameterization and (2) a frozen shower library. Both work by generating the response of the calorimeter to electrons and positrons with Geant 4, and then re-importing the response into the simulation at run-time. In the fast shower parameterization technique, a parameterization is tuned to single electrons and used later by simulation. In the frozen shower technique, actual showers from low-energy particles are imported into the simulation. Simulation in the presence of frozen showers is then required to develop the shower down to ~ 1 GeV, at which point the shower is terminated by substituting a frozen shower. Judicious use of both techniques over the entire electromagnetic portion of the ATLAS calorimeter produces an important improvement of CPU time. We discuss the algorithms and their performance in this talk.
Speaker: Wolfgang Ehrenfeld (Univ. of Hamburg/DESY)
• Grid middleware and tools: GM 2 Carson Hall C

Carson Hall C

Convener: Jeff Templon (NIKEF)
• 165
dCache, the Challenge.
With the start of the Large Hardron Collider at CERN, end of 2007, the associated experiments will feed the major share of their data into the dCache Storage Element technology at most of the Tier I centers and many of the Tier IIs including the larger sites. For a project, not having its center of gravity at CERN, and receiving contributions from various loosely coupled sites in Europe and the US, this is certainly an organizational and financial challenge. The presentation is meant to be the comprehensive overview of the numerous sub-projects collectively known as dCache.org. Beside a brief introduction into the dCache technology itself we will provide an insight into the project structure and the various organizations dCache contributes to and gets contributions by. The technical part will ocus on topics like the SRM 2.2 development, the integration of the improved file name space provider (Chimera), including the ACL plug-in, the pluggable authorization module (gPlazma), the various file location and scheduling mechanisms, the variety of dCache access protocols and many more. Last but not least we are eager to give an outlook on the upcoming short and mid term topics. These are the generic posix data access to dCache via the nfs4.1 protocol, the refurbished tertiary storage connector module, the dCache back-to-back transfer mechanisms and the enhanced meta data storage technology ,provided by chimera.
Speaker: Dr Patrick Fuhrmann (DESY)
• 166
A Distributed Storage System with dCache
The LCG collaboration is encompased by a number of Tier 1 centers. The nordic LCG Tier 1 is in contrast to other Tier 1 centers distributed over most of Scandinavia. A distributed setup was chosen for both political and technical reasons, but also provides a number of unique challenges. dCache is well known and respected as a powerfull distributed storage resource manager, and was chosen for implementing the storage aspects of the nordic Tier 1. In contrast to classic dCache deployments, we deploy dCache over a WAN with limitted bandwith, high latency, frequent network failures, and spanning many administrative domains. These properties provide unique challenges, covering topics such as security, administration, maintenance, upgradability, reliability, and performance. Our initial focus has been on implementing the GridFTP 2 OGF recommendation in dCache and the Globus Toolkit. Compared to GridFTP 1, GridFTP 2 allows for more intelligent data flow between clients and storage pools, thus enabling more efficient use of our limitted bandwith. Future efforts will address other issues, such as reliability in case of network separation.
Speaker: Dr Gerd Behrmann (Nordic Data Grid Facility)
• 167
Tools for the management of stored data and transfer of data: DPM and FTS
As a part of the EGEE project the data management group at CERN has developed and support a number of tools for various aspects of data management: A file catalog (LFC), a key store for encryption keys (Hydra), a grid file access library (GFAL) which transparently uses various byte access protocols to access data in various storage systems, a set of utilities (lcg_utils) for higher level operations on data are all supported. However, in this presentation we will focus on giving an overview of two components in particular: A disc pool manager (DPM) which provides a service to coordinate the storage of files across discs. The DPM features POSIX ACLs on files and pools, file lifetime with garbage collection, optional replication of data within the DPM and authorization based on VOMS grid certificates. The DPM offers an SRM interface, versions 1.1 and 2.2, along with its own control interface. Access to data is supported via gsiftp, rfio and an optional xrootd module is also available. A file transfer service (FTS) allows the replication of data from one data store to another. The FTS features individually configurable, unidirectional management channels. The channels allow allocation of parameters such as number of concurrent transfers, number of parallel streams or TCP buffer size. SRM (version 1.1 or 2.2) is used to send requests to the storage systems. Third party gridftp or SRMCopy initiated transfers are supported.
Speaker: Dr Markus Schulz (CERN)
• 168
Managing ATLAS data on a petabyte-scale with DQ2
The ATLAS detector at CERN's Large Hadron Collider presents data handling requirements on an unprecedented scale. From 2008 on the ATLAS distributed data management system (DQ2) must manage tens of petabytes of event data per year, distributed globally via the LCG, OSG and NDGF computing grids, now known as the WLCG. Since its inception in 2005 DQ2 has continuously managed all datasets for the ATLAS collaboration, which now comprises over 3000 scientists participating from more than 150 universities and laboratories in more than 30 countries. Fulfilling its primary requirement of providing a highly distributed, fault-tolerant as well as scalable architecture DQ2 has now been successfully upgraded from managing data on a terabyte-scale to data on a petabyte-scale. We present improvements and enhancements to DQ2 based on the increasing demands for ATLAS data management. We describe performance issues, architectural changes and implementation decisions, the current state of deployment in test and production as well as anticipated future improvements. Test results presented here show that DQ2 is capable of handling data up to and beyond the requirements of full-scale data-taking.
Speaker: Mr Mario Lassnig (CERN & University of Innsbruck, Austria)
• 169
Role of Digital Forensics in Service Oriented Architectures
Security requirements of service oriented architectures (SOA) are reasonably higher than the classical information technology (IT) architectures. Loose coupling – the inherent benefit of SOA – stipulates security as a service so as to circumvent tight binding of the services. The services integration interfaces are developed with minimal assumptions between the sending and receiving parties. This services aggregation approach is highly beneficial for achieving higher performance level; however, bookkeeping and logging of various events of such dynamic architecture are very complex issues. Security architecture requires these trails of events to establish clearly what happened and why in the post-accident scenario. The techniques employed to determine the reasons of security architecture’s failure to prevent an incident are collectively known as Digital Forensics. It is necessary to develop digital forensics techniques for SOA so that necessary actions can be taken in the wake of a security breach. In this article we explore the role of digital forensics in SOA especially in the mission-critical SOA applications. We envision digital forensics as a sub-service of the security service in SOA. We propose the use of a monitoring service to generate these logs. We then present a mechanism of efficiently managing the logs of various actions based on the lifecycle of these logs. We finally conclude with the open issues and areas for further improvements.
Speaker: Dr Syed Naqvi (CoreGRID Network of Excellence)
• Online computing: OC 1 Oak Bay

Oak Bay

Convener: Brigitte Vachon (McGill University)
• 170
The DZERO Run 2 L3/DAQ System Performance
The DZERO experiment records proton-antiproton collisions at the Fermilab Tevatron collider. The DZERO Level 3 data acquisition (DAQ) system is required to transfer event fragments of approximately 1-20 kilobytes from 63 VME crate sources to any of approximately 240 processing nodes at a rate of 1 kHz. It is built upon a Cisco 6509 Ethernet switch, standard PCs, and commodity VME single board computers (SBCs). We will discuss the running experience of the system since 2002, the incremental upgrades, scaling capabilities, and how physics goals have altered the way we run the system.
Speaker: Prof. Gordon Watts (University of Washington)
• 171
The ATLAS High Level Trigger Steering
The High Level Trigger (HLT) of the ATLAS experiment at the Large Hadron Collider receives events which pass the LVL1 trigger at ~75 kHz and has to reduce the rate to ~200 Hz while retaining the most interesting physics. It is a software trigger and performs the reduction in two stages: the LVL2 trigger should take ~10 ms and the Event Filter (EF) ~1 s. At the heart of the HLT is the Steering software. To minimise processing time and data transfers it implements the novel event selection strategies of seeded, step-wise reconstruction and early rejection. The HLT is seeded by regions of interest identified at LVL1. These and the static configuration determine which algorithms are run to reconstruct event data and test the validity of trigger signatures. The decision to reject the event or continue is based on the valid signatures, taking into account pre-scale and pass-through. After the EF, event classification tags are assigned for streaming purposes. Several powerful new features for commissioning and operation have been added: comprehensive monitoring is now built in to the framework; for validation and debugging, reconstructed data can be written out; the steering is integrated with the new configuration (presented separately), and topological and global triggers have been added. This paper will present details of the final design and its implementation, the principles behind it, and the requirements and constraints it is subject to. The experience gained from technical runs with realistic trigger menus will be described.
Speaker: Dr Simon George (Royal Holloway)
• 172
High Level Trigger Configuration and Handling of Trigger Tables in the CMS Filter Farm
The CMS experiment at the CERN Large Hadron Collider is currently being commissioned and is scheduled to collect the first pp collision data towards the end of 2007. CMS features a two-level trigger system. The Level-1 trigger, based on custom hardware, is designed to reduce the collision rate of 40 MHz to approximately 100 kHz. Data for events accepted by the Level-1 trigger are read out and assembled by an Event Builder through a complex of switched networks. The High Level Trigger (HLT), running on a computing farm consisting of standard CPU (Filter Farm), employs a set of sophisticated software algorithms, based on the same full-fledged reconstruction framework used for offline, to analyze the complete event information, and further reduce the accepted event rate by approximately three orders of magnitude. This paper describes the design and implementation of the HLT configuration management system. The creation of an HLT configuration, consisting of many software modules organized in a number of trigger paths, and its deployment into the distributed online environment consisting of O(1000) CPU, are centered around a robust database design, abstracting the features of the algorithms and their organization in a trigger table. The evolution of the underlying code, and the issues related to migration of existing tables across software releases, are addressed by a thin code parsing layer. The population of tables, using a dedicated GUI, their retrieval by the Run Control System for deployment in the HLT, and access to historic data all use a unique interface. Reformatting and deployment are decoupled from the database, thus permitting the target configuration grammar to evolve independently. Referential integrity and data consistency are expected to be guaranteed by this system across the entire lifetime of the experiment. First experiences from the commissioning of the HLT system are also reported.
Speaker: Emilio Meschi (CERN)
• 173
The LHCb High Level Trigger Software Framework
The High Level Trigger and Data Acquisition system of the LHCb experiment at the CERN Large Hadron Collider must handle proton-proton collisions from beams crossing at 40 MHz. After a hardware-based first level trigger events have to be processed at the rate of 1 MHz and filtered by software-based trigger applications that run in a trigger farm consisting of up to 2000 PCs. The final rate of accepted events is around 2 kHz. This contribution describes the architecture used to host the selection algorithms of the high level trigger on each trigger node, that is based on shared memory event buffers. It illustrates the interplay between event building processes, event filter processes and processes sending accepted events to the storage system. It describes these software components that are based on the Gaudi event processing framework.
Speaker: Dr Markus Frank (CERN)
• 174
The High-Level Trigger at the CMS experiment
The High Level Trigger (HLT) that runs in the 1000 dual-CPU box Filter Farm of the CMS experiment is a set of sophisticated software tools for selecting a very small fraction of interesting events in real time. The coherent tuning of these algorithms to accommodate multiple physics channels is a key issue for CMS, one that literally defines the reach of the experiment's physics program. In this presentation we will discuss studies of the performance of the HLT algorithms for preliminary versions of integrated Trigger Menus.
Speaker: Leonard Apanasevich (University of Chicago at Illinois)
• 175
Event reconstruction algorithms for the ATLAS trigger
The ATLAS experiment under construction at CERN is due to begin operation at the end of 2007. The detector will record the results of proton-proton collisions at a centre-of-mass energy of 14 TeV. The trigger is a three-tier system designed to identify in real-time potentially interesting events that are then saved for detailed offline analysis. The trigger system will select approximately 200 Hz of potentially interesting events out of the 40 MHz bunch-crossing rate (with ~10^9 interactions per second at the nominal luminosity). Algorithms used in the trigger system to identify different event features of interest will be described, as well as their expected performance in terms of selection efficiency, background rejection and computation time per event. The talk will concentrate on recent improvements and on performance studies, using a very detailed simulation of the ATLAS detector and electronics chain that emulates the raw data as it will appear at the input to the trigger system. Checks on the robustness of the algorithms to detector misalignment and miscalibration will also be discussed.
Speaker: Teresa Maria Fonseca Martin (CERN)
• 176
Trigger Selection Software for Beauty physics in ATLAS
The unprecedented rate of beauty production at the LHC will yield high statistics for measurements such as CP violation and Bs oscillation and will provide the opportunity to search for and study very rare decays, such as B→  .The trigger is a vital component for this work and must select events containing the channels of interest from a huge background in order to reduce the 40 MHz bunch crossing rate down to 100-200 Hz for recording, of which only a part will be assigned to B-physics. Requiring a single or di-muon trigger provides the first stage of the B-trigger selection. Track reconstruction is then performed in the Inner Detector, either using the full detector, at initial luminosity, or within Regions of Interest identified by the first level trigger at higher luminosities. Based on invariant mass, combinations of tracks are selected as likely decay products of the channel of interest and secondary vertex fits are performed. Events are selected based on properties such as fit quality and invariant mass. We present fast vertex reconstruction algorithms suitable for use in the second level trigger and event filter (level three). We discuss the selection software and the flexible trigger strategies that will enable ATLAS to pursue a B-physics programme from the first running at a luminosity of about 1031 cm-2s-1 through to the design luminosity running at 1034 cm-2s-1.
Speaker: Dmitry Emeliyanov (RAL)
• Software components, tools and databases: SC 2 Saanich

Saanich

• 177
Development Status and Plans for the LCG Common Database Access Layer (CORAL)
The CORAL package has been developed as part of the LCG Persistency Framework project, to provide the LHC experiments with a single C++ access layer supporting a variety of relational database systems. In the last two years, CORAL has been integrated as database foundation in several LHC experiment frameworks and is used in both offline and online domains. Also, the other LCG Persistency Framework components such as POOL and COOL are now using CORAL to implement their higher-level database operations in a consistent way for all supported database back-ends. This presentation will summarise the CORAL functionality and the experience gained in large-scale physics production activities. We present recent developments, such as support for multi-threaded applications, a python scripting interface and tools for copying data between different databases. Finally, an overview of the remaining development and consolidation activities to prepare for full LHC production will be presented.
Speaker: Dirk Duellmann (CERN)
• 178
COOL Software Development and Service Deployment Status
The COOL project provides software components and tools for the handling of the LHC experiment conditions data. COOL software development is the result of a collaboration between the CERN IT Department and Atlas and LHCb, the two experiments that have chosen it as the base of their conditions database infrastructure. COOL supports persistency for several relational technologies (Oracle, MySQL and SQLite), based on the CORAL Relational Abstraction Layer. For both experiments, Oracle will be the backend used for the deployment of COOL database services at Tier0 (both online and offline) and Tier1 sites. While software development is still ongoing, especially in the area of performance optimization for data insertion and retrieval, the deployment and test of Oracle database services for COOL will be the main focus of the project in 2007. In this presentation, we will review the status and plans of both software development and service deployment of COOL database services at the time of the CHEP conference, just before LHC startup.
Speaker: Marco Clemencic (European Organization for Nuclear Research (CERN))
• 179
LHCb experience with LFC database replication
Database replication is a key topic in the LHC Computing GRID environment to allow processing of data in a distributed environment. In particular LHCb computing model relies on the LHC File Catalog (LFC). LFC is the database catalog which stores informations about files spread across the GRID, their logical names and physical locations of all their replicas. The LHCb computing model requires the LFC to be replicated at Tier1s via Oracle Streams technology. This paper will describe the LFC replicas deployment at Italian INFN National Center for Telematics and Informatics (CNAF) and at other LHCb Tier1 sites and will present subsequent stress test results. The tests were designed to evaluate any delay on the propagation of the streams and the scalability of the system. The tests show the robustness of the replica implementation with performance going beyond the experiment's requirements.
Speaker: Barbara Martelli (Italian INFN National Center for Telematics and Informatics (CNAF))
• 180
Replication and load Balancing strategy of STAR's RDBM
Database demands resulting from offline analysis and production of data at The STAR experiment at Brookhaven National Laboratory's Relativistic Heavy-Ion Collider has steadily increased over the last 6 years of data taking activities. With each year STAR more than doubles events taken with an anticipation of reaching a billion event capabilities as early as next year. The challenges faced from producing and analyzing this magnitude of events have raised issues with regard to distribution of calibrations and geometry data, via databases, to STAR's growing global collaboration. Rapid distribution, availability, ensured synchronization and load balancing have become paramount considerations. Both conventional technology and novel approaches are used in parallel to realize these goals. This paper discusses how STAR uses distribution methods via MySQL master slave replication to distribute its databases; the synchronization issues that arise from this type of distribution and solutions, mostly homegrown, put forth to overcome these issues. Also discussed is a novel approach toward load balancing between slave nodes that assists in maintaining a high availability rate for a veracious community. This load balancing addresses both, pools of nodes internal to given location, as well as balancing the load for remote users between different available locations. Challenges, trade-offs, rationale for decisions and paths forward will be discussed in all cases, presenting a solid production environment with a vision for scalable growth.
Speaker: Mr Michael DePhillips (BROOKHAVEN NATIONAL LABORATORY)
• 181
Integration of the ATLAS Tag Database with Data Management and Analysis Components
The ATLAS Tag Database is an event-level metadata system, designed to allow efficient identification and selection of interesting events for user analysis. By making first-level cuts using queries on a relational database, the size of an analysis input sample could be greatly reduced and thus the time taken for the analysis reduced. Deployment of such a Tag database is underway, but to be most useful it needs to be integrated with the distributed data management (DDM) and distributed analysis (DA) components. This means addressing the issue that the DDM system at ATLAS groups files into datasets for scalability and usability, whereas the Tag database points to events in files. It also means setting up a system which could prepare a list of input events and use both the DDM and DA systems to run a set of jobs. The ATLAS Tag Navigator Tool (TNT) has been developed to address these issues in an integrated way and provide a tool that the average physicist can use. Here, the current status of this work is presented and areas of future work are highlighted.
Speaker: Dr Caitriana Nicholson (University of Glasgow)
• Tuesday, September 4
• Poster 1: Day 2
• Plenary: Plenary 3 Carson Hall

Carson Hall

Convener: Simon Lin (Taiwan)
• 182
Future of Grid Computing
Speaker: Miron Livny (University of Wisconsin)
• 183
HPC at the Petascale and Beyond
IBM's Blue Gene/L system had demonstrated that it is now feasable to run applications at sustained performances of 100's of teraflops. The next generation Blue Gene/P system is designed to scale up to a peak performance of 3.6 Petaflops. This talk will look at some of the key application successes already achieved at the 100TF scale. It will then address the emerging petascale architectures and look at the challanges which arise as the HPC world now starts to consider designing 100 Petaflop and Exaflop Systems. These challanges are very significant and include power, memory bandwidth, network bandwidth, reliability, systems software, and applications.
Speaker: James Sexton (IBM)
• 184
Speaker: Bill St Arnaud (CANARIE)
• 10:00 AM
Coffee Break
• Computer facilities, production grids and networking: CF 3 Carson Hall B

Carson Hall B

Convener: Kors Bos (NIKEF)
• 185
CMS Experiences with Computing Software and Analysis Challenges
In preparation for the start of the experiment, CMS has conducted computing, software, and analysis challenges to demonstrate the functionality, scalability, and useability of the computing and software components. These challenges are designed to validate the CMS distributed computing model by demonstrating the functionality of many components simultaneously. In the challenges CMS has had participation from approximately 40 computing centers and has demonstrated event processing, data transfers, and analysis processing across the globally distributed computing environment. In this presentation, CMS will describe the types of tests performed including the scale achieved from each CMS component and externally provided component, the functionality demonstrated in the test, and the functionality left to validate before the experiment begins. The successes and lessons learned will be summarized and the directions for the future will be outlined.
Speaker: Dr Ian Fisk (FNAL)
• 186
Quattor and QWG Templates : efficient management of (complex) grid sites
Quattor is a tool aimed at efficient management of fabrics with hundred or thousand of Linux machines, still being easy enough to manage smaller clusters. It has been originally developed inside the European Data Grid (EDG) project. It is now in use at more than 30 grid sites running gLite middleware, ranging from small LCG T3 to very large one like CERN. Main goals and specific features of Quattor are : - Abstract, service oriented description of machine configuration, based on “templates” that can be heavily factorized. - Machine configuration described in term of final state, instead of actions required to reach the final state. - Manage both installation and configuration changes from the same configuration description. - Configuration description versioning, allowing very easy rollback of any change. Quattor is particularly well suited for management of complex sites made of several clusters or subsites spread over several geographical locations, like federated T2s. Quattor ability to factorize common part of description configuration and advanced features of PAN language used to do this description allowed to build and maintained a common set of templates that any site can just import and customize without editing them. This resulted in so-called QWG templates, a complete set of standard templates to configure OS and gLite middleware. This results in a very efficient sharing of installation and configuration tasks around the world. Even if Quattor can be and is used to manage non grid resources (including desktops), this talk will concentrate on Quattor benefits to manage grid resources running gLite, especially using QWG templates.
Speaker: Mr Michel Jouvin (LAL / IN2P3)
• 187
Global Grid User Support - Building a worldwide distributed user support infrastructure
The organization and management of the user support in a global e-science computing infrastructure such as EGEE is one of the challenges of the grid. Given the widely distributed nature of the organisation, and the spread of expertise for installing, configuring, managing and troubleshooting the grid middleware services, a standard centralized model could not be deployed in EGEE. This paper presents the model used in EGEE for building a reliable infrastructure for user, virtual organisation and operations support. The model for supporting a production quality infrastructure for scientific applications will be described in detail. The advantages of the chosen model will be presented and the possible difficulties will be discussed. We will describe the ongoing efforts to build a worldwide grid user support infrastructure in the framework of WLCG by achieving interoperability between the EGEE and OSG user support systems. In this paper we will also describe a scheme of how knowledge management can be used in grid user support and first steps towards a realisation in the framework of the EGEE user support infrastructure.
Speaker: Torsten Antoni (Forschungszentrum Karlsruhe)
• 188
Monitoring the EGEE/WLCG Grid Services
Grids have the potential to revolutionise computing by providing ubiquitous, on demand access to computational services and resources. They promise to allow for on demand access and composition of computational services provided by multiple independent sources. Grids can also provide unprecedented levels of parallelism for high-performance applications. On the other hand, grid characteristics, such as high heterogeneity, complexity and distribution create many new technical challenges. Among these technical challenges, failure management is a key area that demands much progress. A recent survey revealed that fault diagnosis is still a major problem for grid users. When a failure appears at the user screen, it becomes very difficult for her to identify whether the problem is in the used application, somewhere in the grid middleware, or even lower in the fabric that comprises the grid. In this paper we present a tool able to check if a given grid service works as expected for a given set of users (Virtual Organisation) on the different resources available on a grid. Our solution deals with grid services as single components that should produce an expected output to a pre-defined input, what is quite similar to unit testing. The tool, called Service Availability Monitoring or SAM, is being currently used by several different Virtual Organizations to monitor more than 300 grid sites belonging to the largest grids available today. We also discuss how this tool is being used by some of those VOs and how it is helping in the operation of the EGEE/WLCG grid.
Speaker: Mr Antonio Retico (CERN)
• Distributed data analysis and information management: DD 3 Saanich

Saanich

Convener: Michael Ernst (BNL)
• 189
The ATLAS Computing Model
The ATLAS Computing Model was constructed after early tests and was captured in the ATLAS Computing TDR in June 2005. Since then, the grid tools and services have evolved and their performance is starting to be understood through large-scale exercises. As real data taking becomes immanent, the computing model continues to evolve, with robustness and reliability being the watchwords for the early deployment. Particular areas of active development are the data placement and data access, and the interaction between the TAGs, the datasets and the Distributed Data Management issues. The earlier high-level policies and models are now being refined into lower level instantiations.
Speaker: Dr Roger Jones (LANCAS)
• 190
CDF experience with Monte Carlo production using LCG Grid
The upgrades of the Tevatron collider and of the CDF detector have considerably increased the demand on computing resources in particular for Monte Carlo production for the CDF experiment. This has forced the collaboration to move beyond the usage of dedicated resources and start exploiting Grid resources. The CDF Analysis Farm (CAF) model has been reimplemented into LcgCAF in order to access Grid resources by using the LCG/EGEE Middleware components. Many sites in Italy and in Europe are accessed via this portal in order to produce Monte Carlo data and in one year of operations we expect about 100,000 Grid jobs submitted by the CDF users. We review here the setup used to submit jobs to Grid sites and retrieve the output, including the Grid components CDF-specific configuration. The batch and interactive monitor tools developed to allow users to verify the jobs status during theirs lifetimes in the Grid environment are described. We analyze the efficiency and typical failure modes of the current Grid infrastructure reporting the performances of different parts of the used system.
Speaker: Dr Simone Pagan Griso (University and INFN Padova)
• 191
ZEUS Grid Usage: Monte Carlo Production and Data Analysis
The detector and collider upgrades for the HERA-II running at DESY have considerably increased the demand on computing resources for the ZEUS experiment. To meet the demand, ZEUS commissioned an automated Monte Carlo(MC) production capable of using Grid resources in November 2004. Since then, more than one billion events have been simulated and reconstructed on the Grid which corresponds to two thirds of the overall MC production in this period. Based on the experience gained in this successful and efficient production on the Grid, a system has been developed for also running analysis jobs on the Grid. It allows standard ZEUS executables to be submitted to the Grid without any changes and with similar commands as used for the batch analysis. With this approach, the users can easily switch to running their analyses on the Grid and benefit from the additional resources without any Grid middleware knowledge. We present the design and implementation of the Monte Carlo production and the analysis systems which are both based on the ZEUS Grid-toolkit. Furthermore, we report on the status of these systems and our experience with large scale production on the Grid.
Speaker: Dr Hartmut Stadie (Universitaet Hamburg)
• 192
BaBar MC Production on the Canadian Grid using a Web Services Approach
The present paper highlights the approach used to design and implement a web services based BaBar Monte Carlo (MC) production grid using Globus Toolkit version 4. The grid integrates the resources of two clusters at the University of Victoria, using the ClassAd mechanism provided by the Condor-G metascheduler. Each cluster uses the Portable Batch System (PBS) as its local resource management system (LRMS). Resource brokering is provided by the Condor matchmaking process, whereby the job and resource attributes are expressed as ClassAds. The important features of the grid are automatic registering of resource ClassAds to the central registry, ClassAds extraction from the registry to the metascheduler for matchmaking, and the incorporation of input/output file staging. Web-based monitoring is employed to track the status of grid resources and the jobs for an efficient operation of the grid. The performance of this new grid for BaBar jobs, and a comparison with the existing Canadian computational grid (Gridx1) based on Globus Toolkit version 2 is presented.
Speaker: Dr Ashok Agarwal (University of Victoria)
• Event processing: EP 3 Carson Hall A

Carson Hall A

Convener: Patricia McBride (Fermilab)
• 193
Software for CMS Reconstruction
At the end of 2007 the first colliding beams from LHC are expected. The CMS Computing model enforces the use of the same software (with different performance settings) for offline and online(HLT) operations; this is particularly true for the reconstruction software: the different settings must allow a processing time per event (typically, numbers for 2x10e33 luminosity are given) of 50 ms at HLT, while 25 sec are allowed for the offline reconstruction. During 2006 CSA06 data challenge the focus has been put on the offline reprocessing. The reconstruction software has substantially improved from the end-of-2005 version, which was able to process only local reconstruction, to a full fledged reconstruction program, with high level objects ready for analysis tasks (electrons, jets, muons, b/tau tagged jets). The same software is also ready to process a non-ideal detector, and takes into account hardware inefficiencies and misalignments. This second mode of operation, which is important for the readiness of data taking, has been tested in CMS slice tests and commissioning tasks, and has shown that the algorithms used for MonteCarlo processeing are well suited to real-world tasks. During 2007, a second data challenge will explore online/offline reconstruction, and will be used as the base for the ready-for-beam demonstration.
Speakers: Prof. Shahram Rahatlou (Univ di Roma La Sapienza) , Dr Tommaso Boccali (INFN Sezione di Pisa)
• 194
Experience with validating GEANT4 v7 and v8 against v6 in BaBar
BaBar Abstract #8 - Track 2 (Event processing) Experience with validating GEANT4 v7 and v8 against v6 in BaBar S. Banerjee, P. Kim, W. Lockman, and D. Wright for the BaBar Computing Group The BaBar experiment at SLAC has been using the GEANT 4 package version 6 for simulation of the detector response to passage of particles through its material. Since 2005 and 2006, respectively, GEANT 4 versions 7 and 8 have been available, providing: improvements in modeling of multiple scattering; corrections to muon ionization and improved MIP signature; widening of the core of electro-magnetic shower shape profiles; newer implementation of elastic scattering for hadronic processes; exact implementation of Bertini cascade models for kaons and lambdas, and updated hadronic cross-sections from calorimeter beam tests. The effects of these changes in simulation are studied in terms of closer agreement of simulation with respect to data distributions of: the hit residuals of tracks in the silicon-vertex tracker; the shower shapes of photons and K_L particles in the electro-magnetic calorimeter; the ratio of energy deposited in the electro-magnetic calorimeter and the flux return of the magnet instrumented with a muon detection system composed of resistive plate chambers, and limited streamer-tubes; and the muon identification efficiency in the muon detector system of the BaBar detector.
Speaker: Swagato Banerjee (University of Victoria)
• Globus BOF Oak Bay

Oak Bay

• 195
Globus BOF
Globus software was devleoped to enable previously disconnected communities to securely share computational resources and data that span organizational boundaries. As a community driven project, the Globus commiunity is continually creating and enhancing Grid technology to make it easier to administer Grids as well as lowering the barriers to entry for both Grid users and Grid developers. In this presentation we will give a brief snapshot of Globus technology today and provide a look into the future of Globus development and how communities are contributing toward an even brighter future tomorrow. Dr. Dan Fraser is the Director of the Community Development & Improvement program for Globus Software, and is also head of the Globus GridFTP team. Formerly he was the Senior Architect for Grid Middleware at Sun Microsystems. He has over a decade of experience in designing and implementing grid solutions. He has a Ph.D. in physics from Utah State University. His current research interests include Service Oriented Science, grid metrics, and in making Grid technologies easier to use and generally more accessible.
Speaker: Dan Fraser (Globus)
• Grid middleware and tools: GM 3 Carson Hall C

Carson Hall C

Convener: Robert Gardner (University of Chicago)
• 196
DIRAC Data Management: consistency, integrity and coherence of data
The DIRAC Data Management System (DMS) relies on both WLCG Data Management services (LCG File Catalogues, Storage Resource Managers and FTS) and LHCb specific components (Bookkeeping Metadata File Catalogue). The complexity of both the DMS and its interactions with numerous WLCG components as well as the instability of facilities concerned, has turned frequently into unexpected problems in data moving and/or data registration, preventing to have a coherent picture of datasets. Several developments in LHCb have been done in order to avoid data corruptions, data missing , data incoherence and inconsistencies among Catalogues and physical storages both through safety measures at data management level (failover mechanisms, check sums, roll back mechanisms) and extensive background checks. In this paper all the tools developed for checking data integrity and consistency will be presented, as well as a Storage Resource Checker, whose aim is to produce an up-to-date accounting of all LHCb storage usage using the LFC mirror database. Goal of this activity is the development of a generic tool suite able to categorize, analyze and systematically cure the disparate problems affecting DM in order to maintain a consistent picture of the main catalogues (Bookkeeping and LFC) and the Storage Elements.
Speaker: Dr Marianne Bargiotti (European Organization for Nuclear Research (CERN))
• 197
Building the WLCG file transfer service
A key feature of WLCG's multi-tier model is a robust and reliable file transfer service that efficiently moves bulk data sets between the various tiers, corresponding to the different stages of production and user analysis. We describe in detail the file transfer service both the tier-0 data export and the inter-tier data transfers, discussing the transition and lessons learned in moving from a reliable software product, the gLite FTS, to a full production service based on that software. The focus is upon the deployment and operational experience of the service gained during the 2006 and 2007 experiment production activities and dress rehearsals. We discuss the software and operational features that have been deployed to meet the reliability and performance needs of the service, and integration of the service with the WLCG and experiment operations.
Speaker: Dr Markus Schulz (CERN)
• 198
glideinWMS - A generic pilot-based Workload Management System
Grids are making it possible for Virtual Organizations (VOs) to run hundreds of thousands of jobs per day. However, the resources are distributed among hundreds of independent Grid sites. A higer level Workload Management System (WMS) is thus necessary. glideinWMS is a pilot-based WMS, inheriting several useful features: 1) Late binding: Pilots are sent to all suitable Grid sites. Only once pilots start are real jobs selected for that resources. No forecasting is needed. 2) Reliability: A broken Grid site will either kill pilot jobs or pilots will detect the problem at startup. Real jobs only start on well-behaved resources. 3) Grid-wide fair share: The relative priorities between jobs of the same VO are set inside the WMS. Grid sites only manage priorities between different VOs. glideinWMS is based on the Condor glidein concept, i.e. a regular Condor pool, with the Condor daemons (startd) being started by pilot jobs. The real jobs are vanilla, standard or MPI universe jobs. glideinWMS is composed of Glidein Factories and VO Frontends, communicating using Condor ClassAds: * Factories publish the available Grid sites, * Frontends match the Grid attributes to job attributes and publish a request for a stream of glideins to suitable Grid sites * Factories pick up the requests and submit the glideins A detailed description of the system will be presented, along with the currently deployed systems inside USCMS production and user analysis frameworks. Integration with frameworks of other VOs will also be presented, as well as the measured scalability limits.
Speaker: Mr Igor Sfiligoi (FNAL)
• 199
Job Submission and Management Through Web Services: the Experience with the CREAM Service
Modern GRID middlewares are built around components providing basic functionality, such as data storage, authentication and security, job management, resource monitoring and reservation. In this paper we describe the Computing Resource Execution and Management (CREAM) service. CREAM provides a Web service-based job execution and management capability for Grid systems; in particular, it is being used within the gLite middleware. CREAM exposes a Web service interface allowing conformant clients to submit and manage computational jobs to a Local Resource Management System. We developed a special component, called ICE (Interface to CREAM Environment) to integrate CREAM in gLite. ICE transfer job submissions and cancellations from the Workload Management System, allowing users to manage CREAM jobs from the gLite User Interface. This paper describes some recent studies aimed at measuring the performance and the reliability of CREAM and ICE, also in comparison with other job submission systems. We discuss recent work towards enhancing CREAM with a BES and JSDL compliant interface.
Speaker: Mr Luigi Zangrando (INFN Padova)
• Software components, tools and databases: SC 3 Lecture

Lecture

Convener: Federico Carminati (CERN)
• 200
Booting ROOT with BOOT
The BOOT project was introduced at CHEP06 and is gradually implemented in the ROOT project. A first phase of the project has consisted in an important restructuring of the ROOT core classes such that only a small subset is required when starting a ROOT application (including user libraries). Thanks to this first phase, the virtual address space required by the interactive version has been reduced by a factor 3. A second phase of the project has the objective to eliminate a substantial fraction of the dictionary code generated by the preprocessor rootcint. This code will be replaced by persistent objects stored in a ROOT file and the CINT stub functions replaced by direct calls to the compiled code. Prototypes have been developped early 2007 and a full implementation is expected at the time of CHEP07. A third phase under prototyping will automatize the autoloading of the code on demand from a central source repository and online compilation of this code with local caches. This will facilitate and speed-up the installation of new versions. A by-product of this phase has been the implementation of a ROOT file cache (presented in another talk) improving drastically the performance of the ROOT I/O in high latency networks. When all these phases will be completed, it should be possible to install and run a ROOT based application from a web browser (a BROOTER).
Speaker: Dr Rene Brun (CERN)
• 201
ROOT Graphics: status and future.
The ROOT graphical libraries provide support for many different functions including basic graphics, high-level visualization techniques, output on files, 3D viewing etc. They use well-known world standards to render graphics on screen, to produce high-quality output files, and to generate images for Web publishing. Many techniques allow visualization of all the basic ROOT data types, projected in different dimensions and coordinate systems and the production of high quality output for publication purposes. This paper will present the current status of the ROOT graphics including recent developments in 2D representations and the latest developments in the 3D area based on OpenGL. As OpenGL is becoming the standard cross-platform basic graphics package, the current work being done to base all ROOT screen rendered graphics (2D and 3D) on OpenGL will also be presented. Finally we will present some visualization techniques not (yet) available in the ROOT framework which might be of some interest in the future.
Speaker: Mr Olivier Couet (CERN)
• 202
Next generation of OpenGL support in ROOT
OpenGL has been promoted to become the main 3D rendering engine of ROOT. This required a major re- modularization of OpenGL support on all levels, from basic window-system specific interface to medium-level object-representation and top-level scene management. This new architecture allows seamless integration of external scene-graph libraries into the ROOT OpenGL viewer as well as inclusion of ROOT 3D scenes into external GUI and OpenGL-based 3D-rendering frameworks. Scene representation was removed from inside of the viewer, allowing scene-data to be shared among several viewers and providing for natural implementation of multi-view canvas layouts. The object-graph traversal infrastructure allows free mixing of 3D and 2D-pad graphics and makes implementation of ROOT canvas in pure OpenGL possible. Scene-elements representing ROOT objects trigger automatic instantiation of user-provided rendering-objects based on the dictionary information and class-naming convention. Additionally, a finer, per- object control over scene-updates is available to the user, allowing overhead-free maintenance of dynamic 3D scenes and creation of complex real-time animations. User-input handling was modularized as well, making it easy to support application-specific scene navigation, selection handling and tool management.
• 203
Improvements in ROOT I/O's Functionality and Performance
For the last several months the main focus of development in the ROOT I/O package has been code consolidation and performance improvements. Access to remote files is affected both by bandwidth and latency. We introduced a pre-fetch mechanism to minimize the number of transactions between client and server and hence reducing the effect of latency. We will review the implementation and how well it works in different conditions (gain of an order of magnitude for remote file access). We will also review new utilities, including a faster implementation of TTree cloning (gain of an order of magnitude), a generic mechanism for object references, and a new entry list mechanism tuned both for small and large number of selections. In addition to reducing the coupling with the core module and becoming its owns library (libRIO) (as part of the general restructuration of the ROOT libraries), the I/O package has been enhanced in the area of XML and SQL support, thread safety, schema evolution, TTreeFormula, and many other areas. We will also discuss various ways, ROOT will be able to benefit from multi-core architecture to improve I/O performances.
Speaker: Mr Philippe Canal (FERMILAB)
• 204
An interface for GEANT4 simulation using ROOT geometry navigation.
The ROOT geometry modeller (TGeo) offers powerful tools for detector geometry description. The package provides several functionalities like: navigation, geometry checking, enhanced visualization, geometry editing GUI and many others, using ROOT I/O. A new interface module g4root was recently developed to take advantage of ROOT geometry navigation optimizations in the context of GEANT4 simulation. The interface can be used either by native GEANT4-based simulation applications or in the more general context of the Virtual Monte Carlo (VMC) framework developed by ALICE offline and ROOT teams. The latter allows running GEANT3, GEANT4 and FLUKA simulations without changing the geometry description nor the user code. The interface was tested and stressed in the context of ALICE simulation framework. A description of the interface, its usage as well as recent results in terms of reliability and performance will be presented. Some benchmarks will be compared for ROOT-TGeo or GEANT4 based navigation.
Speaker: Mr Andrei Gheata (CERN/ISS)
• 12:30 PM
Whale Watching

For further information see:
http://www.chep2007.com/excursions.html

• ISSeG Oak Bay

Oak Bay

• 205
ISSEG: Welcome Oak Bay

Oak Bay

• 206
ISSeG: How to sell Security - and get resources for it Oak Bay

Oak Bay

Effective security needs resources and support from senior management. This session will look at some ways of gaining that support by establishing a common understanding of risk.
• 207
ISSeG: Risk assessments Oak Bay

Oak Bay

This session will look to establish a common understanding of risk and introduce the ISSeG risk assessment questionnaire.
• 4:00 PM
coffee break Oak Bay (Victoria, Canada)

Oak Bay

• 208
ISSeG: Recommendations Oak Bay

Oak Bay

This session will look at some of the emerging recommendations that can be used at sites to improve security.
• 209
ISSeG: Feedback/questions Oak Bay

Oak Bay

• Wednesday, September 5
• Poster 2: Day 1
• 210
A Data Skimming Service for Locally Resident Analysis Data
A Data Skimming Service (DSS) is a site-level service for rapid event filtering and selection from locally resident datasets based on metadata queries to associated "tag" databases. In US ATLAS, we expect most if not all of the AOD-based datasets to be be replicated to each of the five Tier 2 regional facilities in the US Tier 1 "cloud" coordinated by Brookhaven National Laboratory. Entire datasets will consist of on the order of several terabytes of data, and providing easy, quick access to skimmed subsets of these data will be vital to physics working groups. Typically, physicists will be interested in portions of the complete datasets, selected according to event-level attributes (number of jets, missing E_t, etc) and content (specific analysis objects for subsequent processing). In this paper we describe methods used to classify data (metadata tag generation) and to store these results in a local database. Next we discuss a general framework which includes methods for accessing this information, defining skims, specifying event output content, accessing locally available storage through a variety of interfaces (SRM, dCache/dccp, gridftp), accessing remote storage elements as specified, and user job submission tools through local or grid schedulers. The advantages of the DSS are the ability to quickly "browse" datasets and design skims, for example, pre-adjusting cuts to get to a desired skim level with minimal use of compute resources, and to encode these analysis operations in a database for re-analysis and archival purposes. Additionally the framework has provisions to operate autonomously in the event that external, central resources are not available, and to provide, as a reduced package, a minimal skimming service tailored to the needs of small Tier 3 centers or individual users.
Speaker: Marco Mambelli (University of Chicago)
• 211
Since the beginning, one of the design guidelines for the Workload Management System currently included in the gLite middleware was flexibility with respect to the deployment scenario: the WMS has to work correctly and efficiently in any configuration: centralized, decentralized, and in perspective even peer-to-peer. Yet the preferred deployment solution is to concentrate the workload management functionality on a small number of certified hosts. This is certainly favored by system and virtual organization administrators, because it limits the amount of system management needed to provide the service. But it also helps the users of the system because it simplifies the configuration of the user interface. On the negative side, it raises some scalability problems, because the overall system is requested to manage millions of jobs in the not-too-distant future. In this paper we show how a well-known technique, hiding a number of machines under a DNS-based alias mechanism that takes into account suitable load parameters typical of and specific to Workload Management Systems, can be easily applied to the gLite WMS, addressing both the scalability and the usability issues mentioned above and paving the way to more advanced abstraction mechanisms.
Speaker: Marco Cecchi (INFN-CNAF)
• 212
A Job Monitoring System for the LCG Computing Grid
Today, one of the major challenges in science is the processing of large datasets. The LHC experiments will produce an enormous amount of results that are stored in databases or files. These data are processed by a large number of small jobs that read only chunks. Existing job monitoring tools inside the LHC Computing Grid (LCG) provide just limited functionality to the user. These are either command line tools delivering simple text strings for every job or the provided information is very limited. Other tools like GridIce focus on the monitoring of the infrastructure rather than the user application/job. In contrast to these concept, we developed the Python-based "Job execution Monitor". Typically, the first thing to be executed on a worker node is not a binary executable, but a script file which sets up the environment (including environment variables and loading of data from a storage element, a tasks known to be critical). It is the goal of the Job Execution Monitor to monitor the execution of such critical commands and report their success or failure to the user. The core module of the Job Execution Monitor is the script wrapper. To gain detailed information about the job execution, a given script file (bash or python) is executed command by command. After each command, the complete environment is checked and logged. Together with the other components of this system, an expert system tries to classify the reason for a failure. An integration into the Global Grid User Support is planned.
Speaker: Dr Torsten Harenberg (University of Wuppertal)
• 213
A Login Shell interface for INFN-GRID
The user interface is a crucial service to guarantee the Grid accessibility. The goal to achieve, is the implementation of an environment able to hide the grid complexity and offer a familiar interface to the final user. Currently many graphical interfaces have been proposed to simplify the grid access, but the GUI approach appears not very congenital to UNIX developers and users accustomed to work with command line interface. In 2004 the GridShell project proposed an extension of popular UNIX shells such as TCSH and BASH with features supporting Grid computing. Starting from the ideas included in GridShell, we propose IGSH (INFN-GRID SHELL) a new login shell for the INFN-GRID middleware, that interact with the Resource Broker services and integrates in a “naturally way” the grid functionality with a familiar interface. The architecture of IGSH is very simple, it consist of a software layer on the top of the INFN-GRID middleware layer. When some operation is performed by the user, IGSH takes in charge to parse the syntax and translate it in the correspondents INFN-GRID commands according to some semantic rules specified in the next sections. The final user interacts with the underlying distributed infrastructure by using IGSH instead of his default login shell, with the sensation to work on a local machine. Moreover IGSH shows interesting potentialities, by allowing the user to create complex workflow by using the standard shell language.
Speaker: Dr Silvio Pardi (University of Naples `Federico II'' - C.S.I. and INFN)
• 214
A multi-dimensional view on information retrieval of CMS data
The CMS Dataset Bookkeeping System (DBS) search page is a web-based application used by physicists and production managers to find data from the CMS experiment. The main challenge in the design of the system was to map the complex, distributed data model embodied in the DBS and the Data Location Service (DLS) to a simple, intuitive interface consistent with the mental model of physicists analysis the data. We used focus groups and user interviews to establish the required features. The resulting interface addresses the physicist and production manager roles separately, offering both a guided search structured for the common physics use cases as well as a dynamic advanced query interface.
Speaker: Valentin Kuznetsov (Cornell University)
• 215
A new inclusive secondary vertex algorithm for b-jet tagging in ATLAS
A new inclusive secondary vertexing algorithm which exploits the topological structure of weak b- and c-hadron decays inside jets is presented. The primary goal is the application to b-jet tagging. The fragmentation of a b-quark results in a decay chain composed of a secondary vertex from the weakly decaying b-hadron and typically one or more tertiary vertices from c-hadron decays. The decay lengths and charged particle multiplicities involved in these decays, as well as the instrumental resolution, do not allow to separately reconstruct and resolve these vertices efficiently using conventional secondary vertexing algorithms based on the assumption of a common geometrical vertex. These difficulties are partially overcome in the algorithm presented in this paper, that is based on the hypothesis that the primary event vertex and the vertices of the weak b- and c-hadron decays lie on the same line, the flight direction of the b-hadron. The algorithm provides detailed information on the topology of the decay cascade, also allowing the reconstruction of topologies with only one charged particle from a b- and c-hadron decay, respectively, which are difficult to access for conventional algorithms. The algorithm based on this hypothesis is implemented mathematically as an extension of the Kalman Filter formalism for vertex reconstruction and technically as a set of flexible software modules integrated in the ATLAS software framework Athena, which make use of the existing Event Data Model for vertexing and B-Tagging. The application of the algorithm to b-jet tagging and the impact on its performance is shown.
Speaker: Mr Giacinto Piacquadio (Physikalisches Institut - Albert-Ludwigs-Universität Freiburg)
• 216
A novel design approach and new Geant4 physics developments for microdosimetry simulation
Detailed knowledge of the microscopic pattern of energy deposition related to the particle track structure is required to study radiation effects in various domains, like electronics, gaseous detectors or biological systems. The extension of Geant4 physics down to the electronvolt scale requires not only new physics models, but also adequate design technology. For this purpose a novel approach, based on a policy-based class design, has been explored: the usage of this design technique represents an innovative design within Geant4, and more in general for Monte Carlo simulation for particle physics. A policy-based design assembles classes with complex functionality out of simpler ones responsible of a single behavioural or structural aspect. Policies define a class interface or a class template interface; they are more loosely defined than conventional abstract interfaces, as they are syntax oriented rather than signature oriented. A policy-based design is highly customisable; this feature is relevant to cases where a variety of physics models is desirable. It also provides advantages in performance with respect to more conventional techniques for interchangeable algorithms, such as a Strategy pattern: in fact, policies are compile-time bound and are exempt from the drawbacks related to the virtual method table. The novel design and a set of new physics models introduced in Geant4 are presented; this prototype demonstrates the feasibility of the new approach. The new design technique is suitable to be extended to other Geant4 physics domains to improve the flexibility of modelling configuration and the execution performance.
Speaker: Dr Sebastien Incerti (CENBG-IN2P3)
• 217
A pre-identification for electron reconstruction in the CMS particle-flow algorithm.
In the CMS software, a dedicated electron track reconstruction algorithm, based on a Gaussian Sum Filter (GSF), is used. This algorithm is able to follow an electron along its complete path up to the electromagnetic calorimeter, even in the case of a large amount of Bremsstrahlung emission. Because of the significant CPU consumption of this algorithm, however, it can be run only on a limited number of electron candidates. The standard GSF electron track reconstruction is triggered by the presence of high energy isolated electromagnetic clusters. Instead, a pre-identification algorithm based on both the tracker and the calorimeter has recently been developed. It allows electron tracks to be reconstructed within jets with a good efficiency even for small electron transverse momentum. This algorithm as well as its performance in terms of efficiency, mis-identification probability and timing are presented. Its implementation within the particle-flow algorithm is also described.
Speaker: Michele Pioppi (CERN)
• 218
A software and computing prototype for CMS Muon System alignment
A precise alignment of Muon System is one of the requirements to fulfill the CMS expected performance to cover its physics program. A first prototype of the software and computing tools to achieve this goal has been successfully tested during the CSA06, Computing, Software and Analysis Challenge in 2006. Data was exported from Tier-0 to Tier-1 and Tier-2, where the alignment software was run . Re-reconstruction with new geometry files was also performed at remote sites. Performance and validation of the software has also been tested on cosmic data, taken during the MTCC in 2006, Magnet Test Cosmic Challenge.
Speaker: Mr Pablo Martinez (Insitituto de Física de Cantabria)
• 219
ALICE - ARC integration
AliEn or Alice Environment is the Gridware developed and used within the ALICE collaboration for storing and processing data in a distributed manner. ARC (Advanced Resource Connector) is the Grid middleware deployed across the Nordic countries and gluing together the resources within the Nordic Data Grid Facility (NDGF). In this paper we will present our approach to integrate AliEn and ARC, in the sense that ALICE data management and job processing can be carried out on the NDGF infrastructure, using the client tools available in AliEn. The interoperation has two aspects, one is the data management part and the second the job management aspect. The first aspect was solved by using dCache across NDGF to handle data. dCache provides support for several data management tools (among them for xrootd the tools used by AliEn) using the so called "doors". Therefore, we will concentrate on the second part. Solving it, was somewhat cumbersome, mainly due to the different computing models employed by AliEn and ARC. AliEN uses an Agent based pull model while ARC handles jobs through the more "traditional" push model. The solution comes as a module implementing the functionalities necessary to achieve AliEn job submission and management to ARC enabled sites.
Speaker: Dr Josva Kleist (Nordic Data Grid Facility)
• 220
An effective XML based name mapping mechanism within StoRM
In a Grid environment the naming capability allows users to refer to specific data resources in a physical storage system using a high level logical identifier. This logical identifier is typically organized in a file system like structure, a hierarchical tree of names. Storage Resource Manager (SRM) services map the logical identifier to the physical location of data evaluating a set of parameters as the desired quality of services and the VOMS attributes specified in the requests. StoRM is a SRM service developed by INFN and ICTP-EGRID to manage file and space on standard POSIX and high performing parallel and cluster file systems. An upcoming requirement in the Grid data scenario is the orthogonality of the logical name and the physical location of data, in order to refer, with the same identifier, to different copies of data archived in various storage areas with different quality of service. The mapping mechanism proposed in StoRM is based on a XML document that represents the different storage components managed by the service, the storage areas defined by the site administrator, the quality of service they provide and the Virtual Organization that want to use the storage area. An appropriate directory tree is realized in each storage component reflecting the XML namespace schema. In this scenario StoRM is able to identify the physical location of a requested data evaluating the logical identifier and the specified attributes following the XML schema, without querying any database service. This paper presents the namespace schema defined, the mapping mechanism and the technical details of the StoRM implementation.
Speaker: Mr Luca Magnoni (INFN-CNAF)
• 221
An original model for the simulation of the stopping power of negative hadrons
An original model is presented for the simulation of the energy loss of negatively charged hadrons: it calculates the stopping power by regarding the target atoms as an ensemble of quantum harmonic oscillators. This approach allows to account for charge dependent effects in the stopping power, which are relevant at low energy: the differences between the stopping powers of positive and negative charged hadrons may amount to approximately a factor two, which is significant when high precision is required for the evaluation of energy deposit distributions. Related use case are the tails of energy distributions in the development of showers, the environment of radiation background monitors, experiments for antimatter searches, and the recent interest demonstrated for the possible therapeutic applications of antiproton beams. The resulting antiproton stopping powers for different elements are compared against measurements at CERN antiproton experiments and are shown to be in satisfactory agreement with experimental data. The model described is implemented in the Low Energy Electromagnetic package of the Geant4 Toolkit; it represents a significant improvement for the accurate simulation of low energy negative hadrons with respect to previously available models.
Speaker: Stephane Chauvie (INFN Genova)
• 222
An SSH Key management system: easing the pain of managing key/user association
Secure access to computing facilities has been increasingly on demand of practical tools as the world of cyber-security infrastructure has changed the landscape to access control via gatekeepers or gateways. However, the venue of two factor authentication (SSH keys for example) preferred over simpler Unix based login has introduced the challenging task of managing private keys and its association with individual users. Moreover, while a facility could simplify their model as one key one remote user therefore one local user and deploy a strategy along the lines of LDAP-SSH (Darwin project), such approach would not work for facilities allowing mapping between one “real” remote user and many local accounts adding to that the complexity and dimension of possibly multiple servers. We will present an SSH key management system we developed, tested and deployed to address the one to many dilemma in the RHIC/STAR experiment. We will explain its use in an online computing context and explain the problems it addresses amongst which, making possible the management and tracing of group account access spread over many sub-system components (data acquisition, slow control, trigger groups) without the need of publicly known passwords (while keeping track at all times who/where).
Speaker: Dr Jerome Lauret (BROOKHAVEN NATIONAL LABORATORY)
• 223
ATLAS DDM Integration in ARC
The Nordic Data Grid Facility (NDGF) consists of Grid resources running ARC middleware in Scandinavia and other countries. These resources serve many virtual organisations and contribute a large fraction of total worldwide resources for the ATLAS experiment, whose data is distributed and managed by the DQ2 software. Managing ATLAS data within NDGF and between NDGF and other Grids used by ATLAS (the LHC Computing Grid and the Open Science Grid) presents a unique challenge for several reasons. Firstly, the entry point for data, the Tier 1 centre, is physically distributed among heterogeneous resources in several countries and yet must present a single access point for all data stored within the centre. The middleware framework used in NDGF differs significantly from other Grids, specifically in the way that all data movement and registration is performed by services outside the worker node environment. Also, the service used for cataloging the location of data files is different from other Grids but must still be useable by DQ2 and ATLAS users to locate data within NDGF. This paper presents in detail how we solve these issues to allow seamless access worldwide to data within NDGF.
Speaker: Dr Josva Kleist (Nordic Data Grid Facility)
• 224
ATLAS Liquid Argon Calorimeter Reconstruction Software and Commissioning
The ATLAS Liquid Argon Calorimter consists of precision electromagnetic accordion calorimeters in the barrel and endcaps, hadronic calorimeters in the endcaps, and calorimeters in the forward region. The initial high energy collision data at the LHC experiments is expected in the spring of 2008. While tools for the reconstruction of the calorimeter data are quite developed through years of Monte Carlo simulation and test beam studies, the processing, storing and archiving of all peripheral meta-data such as calibration constants, detailed conditions of the sub-detector systems are actively being worked on. The LAr calorimeter consists of over 180,000 electronic channels. The reconstruction of all the signal channels with proper calibration and other conditions is challenging. The current status of these efforts for the ATLAS Liquid Argon Calorimter is presented. The interfaces to access, and techniques to store, the conditions data are introduced. The current effort in commissioning the detector gives invaluable experience in larger scale application of these methods and is discussed in detail.
Speaker: Rolf Seuster (University of Victoria)
• 225
ATLAS Muon Spectrometer Simulation and its Validation Algorithms
The ATLAS detector, currently being installed at CERN, is designed to make precise measurements of 14 TeV proton-proton collisions at the LHC, starting in 2007. Arguably the clearest signatures for new physics, including the Higgs Boson and supersymmetry, will involve the production of isolated final-stated muons. The identification and precise reconstruction of muons are performed using a combination of detector components, including an inner detector, comprising a silicon tracker, pixel detector, and transition radiation tracker, housed in a uniform solenoidal field, and a precision standalone Muon Spectrometer, comprising monitored drift tubes and cathode strip chambers, triggered by resistive plate chambers and thin-gap chambers, and housed in a toroidal field. In order to manage the complexity and to understand the performance of the ATLAS MuonSpectrometer,a detailed full detector simulation is required and it should be kept under control by means of automatic validation procedures. We describe the implementation and the functionalities of the recently developed MuonValidation package, which has been developed as a dedicated tool to monitor and validate the performance of the Full Simulation and Digitization of the Muon System. Its flexible design allows comparisons between different Muon geometrical layouts and different software releases. Validation results based on fully simulated GEANT4 events, using the complete detailed geometrical description of the detector are shown.
Speakers: Dr Daniela Rebuzzi (INFN, Sezione di Pavia) , Dr Nectarios Benekos (Max-Planck-Institut fur Physik)
• 226
Automatic Model Selection Using Machine Learning Techniques for Event Selection in Particle Physics
Advances in statistical learning have placed at our disposal a rich set of classification algorithms (e.g., neural networks, decision trees, Bayesian classifiers, support vector machines, etc.) with little or no guidelines on how to select the analysis technique most appropriate for the task at hand. In this paper we present a new approach for the automatic selection of predictive models based on the characteristics of the data under analysis. According to the particular data distribution, our methodology may decide to choose a learning algorithm able to delineate complex decision boundaries over the variable space (but exhibiting an inevitable high variance), or rather instead use an algorithm less complex that delineates coarse decision boundaries (but exhibiting a desirable low bias). Our experimental analysis looks for the identification of stop1 signal at energy of 1.96 TeV. The problem is inherently difficult because of the existence of background data with identical signal signatures. We report results using several metrics (e.g., accuracy, efficiency), and compare the performance of our methodology to a model produced by a domain expert that separates manually signal events from background events.
Speaker: Dr Ricardo Vilalta (University of Houston)
• 227
Automatic processing of CERN video, audio and photo archives
The digitalization of CERN audio-visual archives, a major task currently in progress, will generate over 40 TB of video, audio and photo files. Storing these files is one issue, but a far more important challenge is to provide long- time coherence of the archive and to make these files available on line with minimum manpower investment. An infrastructure, based on standard CERN services, has been implemented whereby master files, stored in the CERN Distributed File System (DFS), are discovered and scheduled for encoding into lightweight web formats based on predefined profiles. Changes in master files, conversion profiles or in the metadata database (read from CDS, the CERN Document Server) are automatically detected and the media re-encoded whenever necessary. The encoding processes are run on virtual servers provided on-demand by the CERN Server Self Service Center, so that new servers can be easily configured to adapt to higher load. Finally, the generated files are made available from the CERN standard web servers with streaming implemented using Windows Media Services. This paper describes the architecture in detail and analyses its advantages and limitations.
Speaker: Michal Kwiatek (CERN)
• 228
Bulk file transfer and storage with GridSite and SlashGrid
GridSite has extended the industry-standard Apache webserver for use within Grid projects, by adding support for Grid security credentials such as GSI and VOMS. With the addition of the GridHTTP protocol for bulk file transfer via HTTP and the development of a mapping between POSIX filesystem operations and HTTP requests we have extended this scope of GridSite into bulk data transfer and storage. We present measurements of the performance obtained from HTTP over the wide and local area networks and compare this with other available protocols. Finally, we show how the SlashGrid component of GridSite provides transparent access to remove HTTP(S) URLs which applications can use via the operating system's POSIX file access API. SlashGrid/GridSite therefore implement a distributed filesystem based on HTTP(S) and GridSite fileservers, that can be used within a local area network or across the internet. This system solely relies on Grid credentials such as GSI and VOMS and implements a fine-grained access control model without the need to manipulate underlying Unix accounts or permissions.
Speaker: Dr Andrew McNab (University of Manchester)
• 229
CDF Monte Carlo data transfer and storage with Grid tools
The CDF experiment at Fermilab produces Monte Carlo data files using computing resources on both the Open Science Grid (OSG) and LHC Computing Grid (LCG) grids. This data produced must be brought back to Fermilab for archival storage. In the past CDF produced Monte Carlo data on dedicated computer farms through out the world. The data files were copied directly from the worker nodes to a few file servers located at FNAL using rcp and Kerberos authentication. As the experiment has moved from dedicated resources to shared resources on the grid, this technique has proven to be problematic. We plan to show how changing our data delivery model to one of concentrating the data in a few sites located in different regions throughout the world and shipping it back in a coordinated method back to FNAL improved our use of opportunistic CPU cycles. The details of the evaluation process and implemented solution including metrics demonstrating our performance will also be presented.
Speaker: Dr Douglas Benjamin (Duke University)
• 230
CMS CSA06 experience at INFN
The CMS experiment operated a Computing, Software and Analysis Challenge in 2006 (CSA06). This activity is part of the constant work of CMS in computing challenges of increasing complexity to demonstrate the capability to deploy and operate a distributing computing system at the desired scale in 2008. The CSA06 challenge was a 25% exercise, and included several workflow elements: event reconstruction at the CERN Tier-0 center; data distribution to Tier-1's, for archiving and data serving purposes; data skimming, driven by CMS physics groups, and re- reconstruction at Tier-1's; serving of re-processed data to Tier-2's and Grid submission of physics analysis jobs. The INFN region joined this exercise. An overview of the software tools and the computing architecture deployed at INFN for the challenge is given in this report, as well as a summary of the operational experience at INFN-CNAF Tier-1 and INFN Tier-2 centers. A description of the operational experience at INFN during CSA07 challenge - a CMS exercise at a scale of 50% started in July 2007 - are also presented and discussed.
Speaker: Dr Daniele Bonacorsi (INFN-CNAF, Bologna, Italy)
• 231
CMS Tier Structure and Operation of the Experiment-specific Tasks in Germany
In Germany, several university institutes and research centres take part in the CMS experiment. Concerning the data analysis, a couple of computing centres at different Tier levels, ranging from Tier 1 to Tier 3, exists at these places. The German Tier 1 centre GridKa at the research centre at Karlsruhe serves all four LHC experiments as well as for four non-LHC experiments. With respect to the CMS experiment, GridKa is mainly involved in central tasks. The Tier 2 centre in Germany consists of two sites, one at the research centre DESY at Hamburg and one at RWTH Aachen University, forming a federated Tier 2 centre. Both parts cover different aspects of a Tier 2 centre. The German Tier 3 centres are located at the research centre DESY at Hamburg, at RWTH Aachen University, and at the University of Karlsruhe. Furthermore a German user analysis facility is planned to be built. Since the CMS community in German is rather small, a good cooperation between the different sites is essential. This cooperation includes physical topics as well as technical and operational issues. All available communication channels such as email, phone, monthly video conferences, and regular personal meetings are used. For example, the distribution of data sets is coordinated globally within Germany. Also the CMS-specific services such as the data transfer tool PhEDEx or the Monte Carlo production are operated by people from different sites in order to spread the knowledge widely and increase the redundancy in terms of operators.
Speaker: Dr Andreas Nowack (III. Physikalisches Institut (B), RWTH Aachen)
• 232
Complete Distributed Computing Environment for a HEP Experiment: Experience with ARC-Connected Infrastructure for ATLAS
Computing and storage resources connected by the Nordugrid ARC middleware in the Nordic countries, Switzerland and Slovenia are a part of the ATLAS computing grid. This infrastructure is being commissioned with the ongoing ATLAS Monte Carlo simulation production in preparation for the commencement of data taking in late 2007. The unique non-intrusive architecture of ARC, it's straightforward interplay with the ATLAS Production System via the Dulcinea executor, and its performance during the commissioning exercise will be described. ARC support for flexible and powerful end-user analysis within the GANGA distributed analysis framework will be shown. Whereas the storage solution for this grid was earlier based on a large, distributed collection of gridftp-servers, the ATLAS computing design includes a structured SRM-based system with a limited number of storage endpoints. The characteristics, integration and performance of the old and new storage solutions are described. Although the hardware resources in this grid are quite modest, it has provided more than double the agreed contribution to the ATLAS production with an efficiency above 95% during long periods of stable operation. If time allows, the full chain will be demonstrated.
Speaker: Prof. Alexander Read (University of Oslo, Department of Physics)
• 233
Data Analysis through the DIANA Meta-Scheduling Approach
We introduce the concept, design and deployment of the DIANA meta-scheduling approach to solving the challenge of the data analysis being faced by the CERN experiments. The DIANA meta-scheduler supports data intensive bulk scheduling, is network aware and follows a policy centric meta-scheduling that will be explained in some detail. In this paper, we describe a Physics analysis case study using the DIANA meta-scheduler. We demonstrate that a decentralized and dynamic meta-scheduling approach is an effective strategy to cope with the increasing numbers of users, jobs and datasets. We present "quality of service" related statistics for the physics analysis through the application of a policy centric fair-share scheduling model. The DIANA meta-schedulers create a peer-to-peer hierarchy of schedulers to accomplish resource management. They employ scheduling approaches that are able to change with evolving loads and adapt following the dynamic and volatile nature of the resources. The DIANA meta-scheduler also acknowledges the important role of networks in the modern day distributed systems and treats them an equally important resource in the scheduling decisions along with the compute and data resources. This topic formed the contents of a Ph.D. thesis done partly at CERN and partly at the University of the West of England in the United Kingdom. The Physics analysis case study was based on the computing model that was presented in the CMS Technical Design Report. Our results show that Physics analysis scheduling can be made robust by employing fault tolerant and decentralized meta-scheduling.
Speaker: Prof. Richard McClatchey (UWE)
• 234
Data Quality Monitoring and Visualization for the CMS Silicon Strip Tracker
The CMS Silicon Strip Tracker (SST), consisting of more than 10 millions of channels, is organized in about 16,000 detector modules and it is the largest silicon strip tracker ever built for high energy physics experiments. In the first half of 2007 the CMS SST project is facing the important milestone of commissioning and testing a quarter of the entire SST with cosmic muons. The full standard CMS software is deployed for the data acquisition, reconstruction, monitoring and event display. For the first time the detector performance is monitored using an advanced Data Quality Monitoring (DQM) system capable of running on a variety of online and offline environments, in the control room as well as in remote sites. More than 100.000 monitorable quantities are managed by the DQM system that organizes them in a hierarchical structure reflecting the detector arrangement in subcomponents and the various levels of data processing. Monitorable quantities computed at the level of individual detectors are processed to extract automatic quality checks and summary results that can be visualized with specialized graphical user interfaces. In view of the great complexity of the CMS Tracker detector the standard visualization tools based on histograms have been complemented with 2 and 3 dimensional graphical images of the subdetector that can show the whole detector down to single channel resolution. The functionalities of the CMS Silicon Strip Tracker DQM system and the experience acquired during the SST commissioning are discussed here.
Speaker: Dr Domenico Giordano (Dipartimento Interateneo di Fisica)
• 235
dCache NFSv4.1 - true nfs server with SRM interface and HSM support
Starting June 2007, all WLCG data management services have to be ready and prepared to move terabytes of data from CERN to the Tier 1 centers world wide, and from the Tier 1s to their corresponding Tier 2s. Reliable file transfer services, like FTS, on top of the SRM v2.2 protocol are playing a major role in this game. Nevertheless, moving large junks of data is only part of the challenge. As soon as the LHC experiments go online, thousands of physicists across the world will start data analysis, to provide first results as soon as possible. At that point in time, local file access becomes crucial. Currently, large numbers of local file access protocols are supported by various Storage Systems – dcap, gsidcap, rfio-dpm, rfio-castor, http and xrootd. A standard protocol, usable by any unmodified application, assuming POSIX data access, is highly desirable. The NFSv4.1 protocol, defined by IETF and implemented by various Operating System and Storage Box vendors, e.g. EMC, IBM, Linux, NetApp , Panasas and SUN, provides all necessary functionality: security mechanism negotiation (GSS-API, GSI, X509, UNIX), data access protocol negotiation (NFSv4 mandatory), clear distinction between metadata ( namespace ) and data access, support of multiple dataservers, ACLs, client and server crash recovery and much more. The client modules are being developed for AIX, Linux, and the Solaris kernels. NFSv4.1 is an open standard, industry backed protocol which easily integrates into the dCache architecture. Together with the new namespace provider, Chimera, dCache provides a native NFSv4.1 implementation. At the most recent NFS “Bakeathon” at CITI-UMICH September 2006, dCache has proven to be compatible to all existing clients. At the time of this presentation, at least two production level installations are available. The one for the DESY, H1 experiment and for the Tier 2 center at University of Michigan. The official production version will be available as soon as the NFSv4.1 protocol has been released.
Speaker: Mr Tigran Mkrtchyan Mkrtchyan (Deutsches Elektronen-Synchrotron DESY)
• 236
Design Principles of a Web Interface for Monitoring Tools
A monitoring tool for complex Grid systems can gather a huge amount of information that have to be presented to the users in the most comprehensive way. Moreover different types of consumers could be interested in inspecting and analyzing different subsets of data. The main goal in designing a Web interface for the presentation of monitoring information is to organize the huge amount of data in a simple, user-friendly and usable structure. One more problem is to consider different approaches, skills and interests that all the possible categories of users have in looking for the desired information. Starting from the information architecture guidelines for the Web, it is possible to design Web interfaces towards a better user experience and to deal with an advanced user interaction by exploiting the most recent Web standards. In this paper, we present a number of principles for the design of Web interface for monitoring tools that provide a wider, richer range of possibilities for what concerns the user interaction. These principles are based on an extensive review of the current literature in Web design, on the experience with the development of the GridICE monitoring tool and on an analysis of the design choices of other monitoring systems. The described principles can drive the evolution of the Web interface of Grid monitoring tools.
Speakers: Mr Enrico Fattibene (INFN-CNAF, Bologna, Italy) , Mr Giuseppe Misurelli (INFN-CNAF, Bologna, Italy)
• 237
DIRAC Agents and Services
DIRAC Services and Agents are defined in the context of the DIRAC system (the LHCb's Grid Workload and Data Management system), and how they cooperate to build functional sub-systems is presented. How the Services and Agents are built from the low level DIRAC framework tools is described. Practical experiente in the LHCb production system has directed the creation of the current DIRAC framework tools. The framework has been designed to ease the development and deployment of DIRAC components, and to provide a certain level of protection angainst external misuse. Guidelines for service robustness and real-time response used currently in DIRAC will be discussed. Measurements on the current performance of the running DIRAC system from the point of view of the functionality provided by the framework are presented.
Speaker: Dr Ricardo Graciani Diaz (Universidad de Barcelona)
• 238
DIRAC Framework for Distributed Computing
The DIRAC system is made of a number of cooperating Services and Agents that interact between them with a Client-Server architecture. All DIRAC components rely on a low level framework that provides the necessary basic functionality. In the current version of DIRAC these components have been identified as: DISET, the secure communication protocol for remote procedure call and file transfer; Configuration System, providing redundant distributed mechanism for configuration and service discovery; Logging and Monitoring System, a uniform way for all components to report their status and activities, and present that information to the users. The current functionality is the result of the experience collected during the last years of running with DIRAC for LHCb. The design principles and the resulting architecture currently implemented in DIRAC will be presented.
Speaker: Mr Adrian Casajus Ramo (Universitat de Barcelona)
• 239
DIRAC Job Prioritization and Fair Share in LHCb
LHCb accesses Grid through DIRAC, its WorkLoad and Data Management system. In DIRAC all the jobs are stored in central task queues and then pulled onto worker nodes via generic Grid jobs called Pilot Agents. These task queues are characterized by different requirements about CPUtime and destination. Because the whole LHCb community is divided in sets of physicists, developers, production and software managers it is important to assign different priorities to different jobs. A special DIRAC agent is responsible for dynamic calculation of the job priorities. The goals of the job priority agent are to ensure that the whole LHCb community fairly uses and shares the Grid resources and to minimize the overall time spent by the high priority jobs in the DIRAC task queues. In the paper the possible technical approaches to define the job priority are evaluated. In particular, the use of the MAUI scheduler is studied. The results of the application of the proposed job prioritization procedure to real jobs are presented. The current limitations and future work are discussed.
Speaker: Gianluca Castellani (European Organization for Nuclear Research (CERN))
• 240
DIRAC: A community grid solution
The DIRAC system was developed in order to provide a complete solution for using distributed computing resources of the LHCb experiment at CERN for data production and analysis. It allows a concurrent use of over 10K CPUs and 10M file replicas distributed over many tens of sites. The sites can be part of a computing grid such as WLCG or standalone computing clusters all integrated in a single management structure. DIRAC is a generic system with the LHCb specific functionality incorporated through a number of plug-in modules. It can be easily adapted to the needs of other communities. A special attention is paid to the resilience of the DIRAC components to allow an efficient use of non-reliable resources. The DIRAC production management components provide a framework for building highly automated data production systems including data distribution and data driven workload scheduling. In this paper we give an overview of the DIRAC system architecture and design choices. We show how different components are put together to compose an integrated data processing system including all the aspects of the LHCb experiment - from the MC production and raw data reconstruction to the final user analysis.
Speaker: Dr Andrei Tsaregorodtsev (CNRS-IN2P3-CPPM, Marseille)
• 241
DIRAC: Reliable Data Management for LHCb
DIRAC, LHCb’s Grid Workload and Data Management System, utilises WLCG resources and middleware components to perform distributed computing tasks satisfying LHCb’s Computing Model. The Data Management System (DMS) handles data transfer and data access within LHCb. Its scope ranges from the output of the LHCb Online system to Grid-enabled storage for all data types. It supports metadata for these files in replica and bookkeeping catalogues, allowing dataset selection and localisation. The DMS controls the movement of files in a redundant fashion whilst providing utilities for accessing all metadata. To do these tasks effectively the DMS requires complete self integrity between its components and external physical storage. The DMS provides highly redundant management of all LHCb data to leverage available storage resources and to manage transient errors in underlying services. It provides data driven and reliable distribution of files as well as reliable job output upload, utilising VO Boxes at LHCb Tier1 sites to prevent data loss. In this paper the evolution of the DIRAC Data Management System will be presented highlighting successful design choices and limitations discovered.
Speaker: Andrew Cameron Smith (CERN)
• 242
LCG experiments will contain large amount of data in relational databases. Those data will be spread over many sites (Grid or not). Fast and easy access will required not only from the batch processing jobs, but also from the interactive analysis. While many system have been proposed and developed for access to file-based data in the distributed environment, methods of efficient access to relational database resources have been so far largely underdeveloped. Presentation will describe an architecture allowing seamless access to distributed relational data from end-user clients. Implementation is based on the standard protocols and tools, extended with HEP-specific functionality to be easily integrated in LCG frameworks. Major constituents of the architecture are: - Java as the implementation language and JDBC as the databases access protocol - transparent access from other languages using standard interfaces - standardly available Sequoia tool to offer single access point to distributed database resources - specific plugins to accommodate HEP-specific access patterns and provide resource usage optimization via networks of transparent proxy caches - integration in several end-user analysis tools (JAS, jHEPWork,...)
Speaker: Dr Julius Hrivnac (LAL)
• 243
DPM Status and Next Steps
The DPM (Disk Pool Manager) provides a lightweight and scalable managed disk storage system. In this paper, we describe the new features of the DPM. It is integrated in the grid middleware and is compatible with both VOMS and grid proxies. Besides the primary/secondary groups (or roles), the DPM supports ACLs adding more flexibility in setting file permissions. Tools to interact with the DPM at different levels have been extended so that site managers can more dynamically configure and manage their DPM in a consistent way. In addition to rfio and gsiftp, users can now use the xrootd and https protocols to access the DPM. A new version of Storage Resource Manager (SRM) interface, v2.2 has been implemented. One of the novelties is the reserve space concept, useful to guarantee space for a specific user or a group during a given period of time. DPM has been deployed in roughly 80 Tier-2 sites and in several medical institutes. Unlike physics data, medical data is very sensitive. The DPM will offer the possibility to encrypt data throughout the process in a very secure way by implementing a key-distributed system. Performance has been improved by the use of bulk queries. Stressing tests have shown a good robustness of the DPM against concurrent accesses.
• 244
Electron reconstruction in CMS
We describe the strategy developed for electron reconstruction in CMS. Emphasis is put on isolated electrons and on recovering the bremsstrahlung losses due to the presence of the material before the ECAL. Following the strategy used for the high level triggers, a first filtering is obtained building seeds from the clusters reconstructed in the ECAL. A dedicated trajectory building is then used to collect hits up to the ECAL front face. The fit of the electron trajectories involves a Gaussian sum filter, allowing to take into account the bremsstrahlung losses. The different topologies observed in the ECAL and the difference between the momentum at the innermost and outermost track positions are then used to classify electrons according to their patterns regarding radiative losses. These electron classes are finally used to correct the momentum estimate and to improve the electron identification.
Speaker: Mr Claude Charlot (Ecole Polytechnique)
• 245
Enabling a priority-based FairShare in the EGEE Infrastructure
While starting to use the grid in production, applications have begun to demand the implementation of complex policies regarding the use of resources. Some want to divide their users in different priority brackets and classify the resources in different classes, others again content themselves with considering all users and resources equal. Resource managers have to work into enabling these requirements on theri site, in addition to he work necessary to implement policies regarding the use of their resources, to ensure compliance with AUPs. These requirements prescribe the existence of a security framework not only capable of satisfying them, but also flexible enough not to require continuous and unnecessary low-level tweaking of configurations every time the requirement change, and that should also do so in a scalable mode. Anything else would only be detrimental when things are seen from the site administrator point of view. Here we will describe the layout used in several italian sites of the EGEE infrastructure to deal with these requirements, along with a complete rationale of our choices, with the intent of clarifying what issues an administrators may run into when dealing with priority requirements, and what common pitfalls should be avoided at any cost. Beyond the feedback on interfaces for policy management, from VO and site administrators, we will especially report on the aspects coming from the mapping of grid level policies to local computing resource authorization mechanisms at sites like CNAF T1, and how they interfere from management and security point of view.
Speaker: Dr Vincenzo Ciaschini (INFN CNAF)
• 246
Ensuring GRID resource availability with the SAM framework in LHCb
The LHCb experiment has chosen to use the SAM framework (Service Availability Monitoring Environment) provided by the WLCG developers to make extensive tests of the LHCb environment at all the accessible grid resources. The availability and the proper definition of the local Computing and Storage Elements, user interfaces as well as the WLCG software environment are checked. The same framework is also used to pre-install the LHCb applications in the shared software area provided by each site. The deployment of the LHCb applications is based on a python tool developed inside the experiment. It is used for software management including incremental installation of interdependent packages and clean package removal. The software installation is followed by a full application run tests. According to the results of the experiment specific SAM tests, the sites are allowed to be used in the LHCb production system called DIRAC. The possibility of automated dynamic site certification using the SAM test suite is explored. This paper will describe the various ways of the LHCb use of the SAM framework. Practical experience in the recent production runs, current limitations and future developments will be presented.
Speaker: Mr Joel Closier (CERN)
• 247
Event reconstruction in the Forward and Backward Silicon detectors of HERA experiment H1
Stand-alone event reconstruction was developed for the Forward and the Backward Silicon Trackers of the H1 experiment at HERA. The reconstruction module includes the pattern recognition algorithm, a track fitter and primary vertex finder. The reconstruction algorithm shows high efficiency and speed. The detector alignment was performed to within an accuracy of 10 um which corresponds to the spatial hit resolution. The reconstruction software has been used for on-line and off-line event analysis of the silicon strip detector's data.
Speakers: Mr Sergey Gorbunov (GSI) , Dr alexander glazov (DESY)
• 248
Experience with xrootd/PROOF solution as a data access model for a Computing Center
We present our experience in setting up an xrootd storage cluster at CC-IN2P3 - a LCG Tier-1 computing Center. The solution consists of xrootd storage cluster made of NAS boxes and includes an interface to dCache/SRM, and Mass Storage System. The feature of this system is integration of PROOF for facilitation of analysis. The setup allows to take advantage of ease of administrative burden, scalability and performance of xrootd, reduction of total cost due to integration of PROOF, while satisfying requirements for SRM interface, strong authentication for transfer and complex data management applied by LCG to site's storage. The same setup can be used at large or small HEP computing centers as analysis facility and data access systems. Variations of this setup for smaller sites are also considered. Feedback from experiments on this setup is also reported.
Speaker: Mr Trunov Artem (CC-IN2P3 (Lyon) and EKP (Karlsruhe))
• 249
Experimental Evaluation of Job Provenance in ATLAS environment
Grid middleware stacks, including gLite, matured into the state of being able to process upto millions of jobs per day. Logging and Bookkeeping, the gLite job-tracking service keeps pace with this rate, however it is not designed to provide a long-term archive of executed jobs. ATLAS---representative of large user community--- addresses this issue with its own job catalogue (prodDB). Development of such a customized service took considerable effort which is not easily affordable by smaller communities and is not easily reused. On the contrary, Job Provenance (JP) is a generic gLite service designed for long-term archive of information on executed jobs. Its design priorities are: (i) scalability -- store data on billions of jobs; (ii) extensibility -- virtually any data format can be uploaded and handled by plugins; (iii) uniform data view -- all data are logically transformed into RDF-like data model, using appropriate namespaces to avoid ambiguities; (iv) configurability -- highly customizable components maintaining pre-cooked queries provide efficient query interface. We present first results of experimental JP deployment for the ATLAS production infrastructure. JP installation was fed with a part of ATLAS production jobs (thousands of jobs per day). We provide a functional comparison of JP and ATLAS prodDB, discuss reliability, performance and scalability issues, and focus on the application level functionality as opposed to pure Grid middleware functions. The main outcome of this work is a demonstration that JP can complement large-scale application-specific job catalogue services, as well as serve similar purpose where these are not available.
Speaker: Ludek Matyska (CESNET)
• 250
Exploring Capable Systems for Fast Data Distribution over End-to-End 10Gb/s Paths
A primary goal of the NSF-funded UltraLight Project is to expand existing data-intensive grid computing infrastructures to the next level by enabling a managed network that provides dynamically constructed end-to-end paths (optically or virtually, in whole or in part). Network bandwidth used to be the primary limiting factor, but with the recent advent of 10Gb/s network paths end-to-end, the end system has become the bottleneck for fast data distribution. As an additional goal of UltraLight we have been exploring tradeoff issues with relatively inexpensive solutions for capable end systems. The candidate system should be capable of driving 10 Gb/s WAN links in order to provide fast data cache capabilities for LHC(Large Hadron Collider) computing model. In this paper, we perform various synthetic and application benchmarks on each architectural component in the data path of a disk-to-disk transfer. First, we find that disk subsystems are usually the main limiting factor for fast data distribution. Therefore, the candidate platforms are required to provide wide I/O paths to accommodate high-performance I/O devices. Second, we investigate a broad range of tunable parameters in the operating system and their impact on the throughput of disk-to-disk transfer. We observe improvement exceeding a factor of 2 in disk I/O throughput which directly impacts the throughput of large disk-to-disk transfers. We also find that disk and memory access patterns together with the size of various buffers play critical roles in maximizing the throughput of disk-to-disk transfers.
Speaker: Mr Kyu Park (Department of Electrical and Computer Engineering, University of Florida)
• 251
Extension of the DIRAC workload-management system to allow use of distributed Windows resources
The DIRAC workload-management system of the LHCb experiment allows coordinated use of globally distributed computing power and data storage. The system was initially deployed only on Linux platforms, where it has been used very successfully both for collaboration-wide production activities and for single- user physics studies. To increase the resources available to LHCb, DIRAC has been extended so that it also allows use of Microsoft Windows machines. As DIRAC is mostly written in Python, a large part of the code base was already platform independent, but Windows-specific solutions have had to be found in areas such as certificate-based authentication and secure file transfers, where .NetGridFTP has been used. In addition, new code has been written to deal with the way that jobs are run and monitored under Windows, enabling interaction with Microsoft Windows Compute Cluster Server 2003 on sets of machines where this is available. The result is a system that allows users’ transparent access to Linux and Windows distributed resources. This paper gives details of the Windows-specific developments for DIRAC, outlines the experience gained deploying the system at a number of sites, and reports on the performance achieved running the LHCb data-processing applications.
Speaker: Ms Ying Ying Li (University of Cambridge)
• 252
Fast Simulations for the PANDA Experiment at FAIR
As one of the primary experiments to be located at the new Facility for Antiproton and Ion Research in Darmstadt the PANDA experiment aims for high quality hadron spectroscopy from antiproton proton collisions. The versatile and comprehensive projected physics program requires an elaborate detector design. The detector for the PANDA experiment will be a very complex machine consisting of a large number of different subdetectors like various tracking detectors, electromagnetic calorimeters, different devices for particle identification etc. The simulation of such a system is a very demanding task in terms of computing power as well as man-power, since in this early stage of the experiment the detector design is not completely fixed and different options for the realization of various subsystems have to be taken into account. For optimization and tuning of the individual detector components a reliable simulation is mandatory, which allows to investigate acceptance and efficiency of various physics benchmark channels for large numbers of events. Therefore, a fast simulation providing effective parametrization of detector acceptance and resolution was implemented. Presented will be the techniques for various aspects of this simulation model as well as applications.
Speaker: Dr Klaus Goetzen (GSI Darmstadt)
• 253
Fluka and Geant4 simulations using common geometry source and digitization algorithms
Based on the ATLAS TileCal 2002 test-beam setup example, we present here the technical, software aspects of a possible solution to the problem of using two different simulation engines, like Geant4 and Fluka, with the common geometry and digitization code. The specific use case we discuss here, which is probably the most common one, is when the Geant4 application is already implemented. Our goal then is to run the same simulation using the Fluka package by re-using the maximum number of the existing components. For simple setups, a tool (FLUGG) already exists that allows to use the Fluka engine while the navigation is performed with Geant4, starting from a description of the geometry in terms of Geant4 classes. In complex applications, however, the geometry is often built up at run time from the information stored in a database, and in these cases such a tool cannot be used directly; furthermore, it does not deal with sensitive detectors and digitization. We show how it is possible to overcome these two problems by building around FLUGG a set of tools for reading common Geometry Description Markup Language (GDML) files as well as for generating the output in the format allowing common processing algorithms.
Speaker: Dr Manuel Venancio Gallas Torreira (CERN)
• 254
Full-scale CMS Tracker application of the Kalman Alignment Algorithm
The Kalman alignment algorithm (KAA) has been specifically developed to cope with the demands that arise from the specifications of the CMS Tracker. The algorithmic concept is based on the Kalman filter formalism and is designed to avoid the inversion of large matrices. Most notably, the KAA strikes a balance between conventional global and local track-based alignment algorithms, by restricting the computation of alignment parameters not only to alignable objects hit by the same track, but also to all other alignable objects that are significantly correlated. Nevertheless, this feature also comes with various trade-offs: Mechanisms are needed that affect which alignable objects are significantly correlated and keep track of these correlations. Due to the large amount of alignable objects involved at each update (at least compared to local alignment algorithms), the time spent for retrieving and writing alignment parameters as well as the required user memory (RAM) becomes a significant factor. The full-scale test presented here, i.e., the employment of the KAA to the (misaligned) CMS Tracker, demonstrates the feasability of the algorithm in a realistic scenario. It is shown that both the computation time and the amount of required user memory are within reasonable bounds, given the available computing resources, and that the obtained results are satisfactory.
Speaker: Mr Edmund Widl (Institut für Hochenergiephysik (HEPHY Vienna))
• 255
GFAL and LCG-Util
GFAL, or Grid File Access Library, is a C library developed by LCG to give a uniform POSIX interface to local and remote Storage Elements on the Grid. LCG-Util is a set of tools to copy/replicate/delete files and register them in a Grid File Catalog. In order to match experiment requirements, these two components had to evolve. Thus, the new Storage Resource Manager interface, SRM v2.2, is now supported. In addition to that, important requirements were to have a python interface to GFAL/LCG-Util, and to fully support Logical File Name (LFN) at GFAL level. Data privacy is a very important issue for some people. Therefore, we have to integrate Hydra client into GFAL. It allows to manage encrypted files and the cryptographic keys. Others important topics of development are to optimize the number of requests to BDII, and also to provide Perl API to GFAL and LCG-Util.
Speaker: Remi Mollon (CERN)
• 256
gPlazma in dCache
gPlazma is the authorization mechanism for the distributed storage system dCache. Clients are authorized based on a grid proxy and may be allowed various privileges based on a role contained in the proxy. Multiple authorization mechanisms may be deployed through gPlazma, such as legacy dcache-kpwd, grid-mapfile, grid-vorolemap, or GUMS. Site-authorization through SAZ is also supported. Services within dCache requesting authorization contact gPlazma through the dCache cell mechanism and recieve a mapping of user credentials and a set of obligations which define the user's privilege.
Speaker: Ted Hesselroth (Fermi National Accelerator Laboratory)
• 257
Grid Information system for EGEE, scalability and performance assessment and plans
Grid Information Systems are mission-critical components for production grid infrastructures. They provide detailed information which is needed for the optimal distribution of jobs, data management and overall monitoring of the Grid. As the number of sites within these infrastructure continues to grow, it must be understood if the current systems have the capacity to handle the extra load. EGEE is the worlds largest production grid infrastructure and hence puts the greatest demands on its information system. Currently 230 sites publish 23 MBytes every 2 minutes into the system and single top level nodes have seen 2 million queries per day. We expect this to grow in the next two years by a factor 5 to 10. This paper describes the current requirements for the EGEE information system obtained from monitoring the usage patterns. From this data, a number of test conditions are derived for determining the subsequent performance and scalability of a grid information system. The tests are then applied to the BDII, which is the information system implementation used by the EGEE infrastructure. The results are given and show the limits of the underlying OpenLDAP Server implementation used within BDII. As a consequence, new techniques and concepts for enhancing the performance are suggested.
Speaker: Mr Laurence Field (CERN)
• 258
GSIMF: A Grid Software Installation Management Framework
To process the vast amount of data from high energy physics experiments, physicists rely on Computational and Data Grids; yet, the distribution, installation, and updating of a myriad of different versions of different programs over the Grid environment is complicated, time-consuming, and error-prone. We report on the development of a Grid Software Installation Management Framework (GSIMF) for managing versioned files with software applications and file-based databases over the Grid infrastructure. A set of Grid services and tools automates the software and databases installation management process by installing and removing software packages on behalf of users. We have developed key prototype Grid services for querying available software packages and installing software on distributed Grid computing elements, demonstrated the prototype services using an ATLAS analysis software package, and explored and investigated other software and databases installation management strategies. We report on hardening the implementations and enhancing the functionality of the prototype Grid services and studies of the applicability of GSIMF for the installation and management of database releases for the ATLAS experiment at the LHC. The new Grid management framework should enable users to remotely install programs and tap into the computing power provided by Grids. The new management framework should find use in various data intensive and collaborative applications, such as nuclear physics experiments, space science observations, and climate modeling.
Speaker: Alexandre Vaniachine (Argonne National Laboratory)
• 259
HEP System Management Working Group and WEB site
System Management Working Group (SMWG) of sys admins from Hepix and grid sites has been setup to address the fabric management problems that HEP sites might have. The group is open and its goal is not to implement new tools but to share what is already in use at sites according to existing best practices. Some sites are already publicly sharing their tools and sensors and some other sites do write very good documentation and share it. The aim is to extend this to a general practice and in a more organised way and avoid the duplication of effort that occurs when system administrators are solving mostly the same problems over and over. The result has been the creation of a WEB site (www.sysadmin.hep.ac.uk) that hosts a subversion repository for management and monitoring tools and a wiki. It works as a file sharing system and single entry point for documentation distributed in other sites. The site, based on gridsite. This paper describes how the group is working and what has been achieved so far.
Speaker: Ms Alessandra Forti (University of Manchester)
• 260
High Performance Data Analysis for Particle Physics using the Gfarm File System
The Belle experiment operates at the KEKB accelerator, a high luminosity asymmetric energy e+ e- collider. The Belle collaboration studies CP violation in decays of B meson to answer one of the fundamental questions of Nature, the matter-anti-matter asymmetry. Currently, Belle accumulates more than one million B Bbar meson pairs that correspond to about 1.2 TB of raw data in one day. The amount of raw data is expected to increase by 50 times after an upgrade of the KEKB accelerator. The challenge is how to realize required high performance data access and scalable data computing. Our solution is the Gfarm file system. It is a commodity-based Grid-wide network shared file system that federates local storage of cluster nodes; moreover it provides scalable I/O performance with distributed data access. We constructed a Gfarm file system with 26 TB capacity and 52 GB/sec I/O bandwidth, integrating local disks of 1112 compute nodes in the KEKB computing facility, and measured scalability of disk I/O performance up to more than 1000 nodes. We also performed a real Belle data analysis program using more than 700 nodes at the speed of 24GB/s, reducing the Belle data analysis time by a factor of about 1,000.
Speaker: Prof. Nobuhiko Katayama (High Energy Accelerator Research Organization)
• 261
High precision physics simulation: experimental validation of Geant4 Atomic Relaxation
A component of the Geant4 toolkit is responsible for the simulation of atomic relaxation: it is part of a modelling approach of electromagnetic interactions that takes into account the detailed atomic structure of matter, by describing particle interactions at the level of the atomic shells of the target material. The accuracy of Geant4 Atomic Relaxation has been evaluated against the experimental measurements of the NIST Physical Reference Data, which include a systematic review of the experimental body of knowledge collected and evaluated over several decades of experimental activity. The validation study concerns X-ray and Auger transition energies. The comparison of the simulated and experimental data with rigorous statistical methods demonstrates the excellent accuracy of the Geant4 simulation models; precision better than 0.5 % is achieved in most cases. The results of this validation study are relevant to various experimental fields, both for elemental analysis studies and for precise simulation of energy deposit distributions; they are also important for the design and optimization of novel tracking detectors based on nanotechnologies, which are sensitive to the effects of charged particles travelling short path lengths, like Auger electrons.
Speaker: Alfonso Mantero (INFN Genova)
• 262
Implementation of chamber mis-alignments and deformations in the ATLAS Muon Spectrometer description and estimate of the muon reconstruction performance reconstruction performance
The Atlas Muon Spectrometer is designed to reach a very high transverse momentum resolution for muons in a pT range extending from 6 GeV/c up to 1 Tev/c. The most demanding design goal is an overall uncertainty of 50 microns on the sagitta of a muon with pT = 1 TeV/c. Such precision requires an accurate control of the positions of the muon detectors and of their movements during the experiment operation. Moreover, the light structure of the Muon Spectrometer, consisting mainly of drift tubes assembled in three layers of stations, imply sizable distortions of the nominal layout of individual chambers, due to mechanical stress and thermal gradients. Corrections for mis-alignments and deformations, which will be provided run-time by an optical alignment system, must be integrated in the software chain leading to track reconstruction and momentum measurement. Here we discuss the implementation of run-time dependent corrections for alignment and distortions in the detector description of the Muon Spectrometer along with the strategies for studying such effects in dedicated simulations. Some preliminary results obtained in the context of the ATLAS Condition Data Challenge effort are also presented.
Speaker: Dr Daniela Rebuzzi (INFN Pavia and Pavia University)
• 263
Implementation of INCL4 cascade with ABLA evaporation in Geant4
We introduce a new implementation of Liege cascade INCL4 with ABLA evaporation in Geant4. INCL4 treats hadron, Deuterium, Tritium, and Helium beams up to 3 GeV energy, while ABLA provides treatment for light evaporation residues. The physics models in INCL4 and ABLA and are reviewd with focus on recent additions. Implementation details, such as first version of object oriented design, are presented, and C++ performance is compared with original FORTRAN implementation. Testing framework validating the FORTRAN-C++ translation is based on ROOT software. Some of the advanced features in testing environment, such as ROOT scripting and automatic documentation features are explained. In addition, we introduce a new Geant4 example application using INCL4 and ABLA models, that demonstrate the physics and compare results against other models previously made available in Geant4, such as Bertini cascade. Finally we outline the future development of Liege cascade, INCL5, in the context of Geant4 hadronic physics framework.
Speaker: Aatos Heikkinen (Helsinki Institute of Physics, HIP)
• 264
Implementing SRM V2.2 Functionality in dCache
The Storage Resource Manager (SRM) and WLCG collaborations recently defined version 2.2 of the SRM protocol, with the goal of satisfying the requirement of the LCH experiments. The dCache team has now finished the implementation of all SRM v2.2 elements required by the WLCG. The new functions include space reservation, more advanced data transfer, and new namespace and permission functions. Implementation of these features required an update of the dCache architecture and evolution of the services and core components of dCache Storage System. Implementation of SRM Space Reservation led to new functionality in the Pool Manager and the development of the new Space Manager component of dCache, responsible for accounting, reservation and distribution of the storage space in dCache. SRM's "Bring Online" function required redevelopment of the Pin Manager service, responsible for staging files from the back-end tape storage system and keeping these files on disk for the duration of the Online state. The new SRM concepts of AccessLatency and RetentionPolicy led to the definition of new dCache file attributes and new dCache pool code that implements these abstractions. SRM permission management functions led to the development of the Access Control List support in the new dCache namespace service, Chimera. I will discuss these new features and services in dCache, provide motivation for particular architectural decisions and describe their benefits to the Grid Storage Community.
Speaker: Timur Perelmutov (FERMI NATIONAL ACCELERATOR LABORATORY)
• 265
Integration of the ATLAS VOMS system with the ATLAS Metadata Interface
AMI is an application which stores and allows access to dataset metadata for the ATLAS experiment. It provides a set of generic tools for managing database applications. It has a three-tier architecture with a core that supports a connection to any RDBMS using JDBC and SQL. The middle layer assumes that the databases have an AMI compliant self-describing structure. It provides a generic web interface and a generic command line interface. A Virtual Organisation Membership Service (VOMS) is an authorisation system for Virtual Organisations (VO's). The ATLAS VO has a VOMS system which contains its own authorisation information. This presentation provides an account of the development of a Java based solution to integrate the ATLAS VOMS system to the ATLAS Metadata Interface (AMI). The prerequisites authentication and authorisation demand on grid architecture are explained. The current workings of a VOMS system and the resulting requirements this has on a client of this system are discussed. We explore possible solutions to these requirements before detailing the mechanism behind the chosen solution and how it was integrated with the AMI framework.
Speaker: Mr Thomas Doherty (University of Glasgow)
• 266
Mapping GRID to Site Credentials using GUMS
Identity mapping is necessary when a site's resources do not use GRID credentials natively, but instead use a different mechanism to identify users, such as UNIX accounts or Kerberos principals. In these cases, the GRID credential for each incoming job must be associated with an appropriate site credential. Many sites consist of a heterogeneous environment with multiple gatekeepers, which can make control and security difficult. Therefore, a single site-wide usage policy is desirable. GUMS (Grid User Management System) is such a Grid Identity Mapping Service providing a single site-wide usage policy. It was designed to integrate with the site's local information services (such as HR databases or LDAP) and with GRID information services (such as VOMS). When a request comes in to a gatekeeper, the gatekeeper contacts the GUMS server to verify a user has appropriate membership (via VOMS, LDAP, or manual user groups), and if so, retrieves the mapping to the local identity (via shared group accounts, a pool of accounts, or a manually defined account). GUMS supports extended X509 certificates that include group membership and role, which influence the mappings. It provides a web interface for managing and testing the server as well as a command line tool. It is contained in the OSG RBAC infrastructure. GUMS has been improved to include a more comprehensive web interface, and we plan to implement recyclable accounts to ensure scaling. GUMS is well suited to sites needing secure and centrally managed GRID to site credential mapping capabilities.
Speaker: Mr Jay Packard (BNL)
• 267
Medical Data Management Status and Plans
The goal of the Medical Data Management (MDM) task is to provide secure (encrypted and under access control) access to medical images, which are stored at hospitals in DICOM servers or are replicated to standard grid Storage Elements (SE) elsewhere. In gLite 3.0 there are three major components to satisfy the requirements: The dCache/DICOM SE is a special SE, which encrypts every requested image with a file specific key. It does not provide a storage area on its own, but interfaces a hospital's DICOM server to the grid. The gLite I/O server with a Fireman catalog service provides the access control by wrapping an SE, which holds medical images. And finally Hydra client library does the en/decryption of the files, using the file specific keys stored in the Hydra keystore. In gLite R3.1 we are planning to simplify the software stack by relying on richer functionality of the underlying components: as storage elements (for example DPM) provide ACLs on individual files, we can remove the wrapping gLite I/O layer from a storage element and access it directly from the client side. Refactoring of the dCache/DICOM SE is also necessary to unify the server side en/decryption and access control functionality in a single component. Finally the Hydra keystore is being split into distributed services for reliability and to reduce the impact of a compromised key server.
Speaker: Akos Frohner (CERN)
• 268
Modular grid middleware configuration system
Configuration is an essential part of the deployment process of any software product. In the case of Grid middleware the variety and complexity of grid services coupled with multiple deployment scenarios make the provision of a coherent configuration both more important and more difficult. The configuration system must provide a simple interface which strikes a balance between the requirements of small university laboratories and those of large computing centers. It should also be simple to maintain alongside rapid developments in the software it is intended to configure. This paper describes the evolution of a modular configuration system (YAIM) which is used to configure different Grid middleware products. It summarizes the lessons learned during several years of LCG/gLite middleware production use and presents a new approach chosen to face the forthcoming challenges. Along with the main design considerations and implementations decisions, an example of a fully integrated third party grid service (dCache) is also discussed.
Speaker: Dr Robert Harakaly (CERN)
• 269
MonALISA: An Agent Based, Dynamic Service System to Monitor, Control and Optimize Distributed Systems
MonaLISA (Monitoring Agents in A Large Integrated Services Architecture) provides a distributed service for monitoring, control and global optimization of complex systems including the grids and networks used by the LHC experiments. MonALISA is based on an ensemble of autonomous multi-threaded, agent-based subsystems which able to collaborate and cooperate to perform a wide range of monitoring and decision tasks in large scale distributed applications, and to be discovered and used by other services or clients that require such information. It is a fully distributed system with no single point of failure. The system is deployed now at more than 340 sites, serving several large Grid communities (ALICE, CMS, OSG, Ultralight, LCG-Russia… ), and it is monitoring around one million parameters in near real-time (complete information for computing nodes, jobs, end to end connectivity, accounting, different grid services, network traffic and topology). MonALISA and its APIs are currently used by different tools in High Energy Physics (CMS job submission systems, Alien, Xrootd, Ganga, Diane…) to collect specific monitoring data which is used as an automatic feedback to different user communities to understand how these complex systems are used and to detect problems. The system is able to react to specific conditions, triggered by alarm conditions, and thus to automatically select an appropriate action. MonALSIA is also used for optimizing global workflows in distributed systems.
Speaker: Dr Iosif Legrand (CALTECH)
• 270
Monitoring of Grid Behaviour by LHCb
Facilities offered by WLCG are extensively used by LHCb in all aspects of their computing activity. A real time knowledge of the status of all Grid components involved is needed to optimize their exploitation. This is achieved by employing different monitoring services each one supplying a specific overview of the system. SAME tests are used in LHCb for monitoring the status of CE services hence its availability for LHCb; DIRAC proprietary monitoring tools allow to track experiment application problems; a Dirac transfer monitoring system reports about failures in data transfers; FTS transfer tests are also closely monitored and the ARDA Dashboard provides complementary information on site reliability. This paper addresses the lack of correlation and analysis of all the available information and proposes a solution for such a central service.
Speaker: Gianluca Castellani (CERN)
• 271
Monitoring with MonAMI: a case study.
Computing resources in HEP are increasingly delivered utilising grid technologies, which presents new challenges in terms of monitoring. Monitoring involves the flow of information between different communities: the various resource-providers and the different user communities. The challenge is providing information so everyone can find what they need: from the local site administrators, regional operational centres through to end-users. To meet this challenge, MonAMI was developed. MonAMI aims to be a universal sensor framework with a plugin architecture. This plugin structure allows flexibility in what is monitored and how the gathered information is reported. MonAMI supports gathering statistics from services like MySQL and Apache. The gathered data can be sent to many monitoring systems, including Ganglia, Nagios, MonaLisa and R-GMA. This flexibility allows MonAMI to be plumbed into whatever monitoring system is being used. This avoids the current duplication of sensors, allowing gathered statistics to be presented within a greater diversity of monitoring systems. Using the MonAMI framework, sensors have been developed for the DPM and dCache storage systems, both common at HEP grid centres. The development of these tools specifically to tackle the challenges of high availability storage is described. We illustrate how MonAMI's architecture allows a single sensor to both deliver long term trend information and to trigger alarms in abnormal conditions. The use of MonAMI within the ScotGrid distributed Tier-2 is examined as a case study, illustrating both the ease with which MonAMI integrates into existing systems and helps provide a framework for extending monitoring where necessary.
Speaker: Dr Paul Millar (GridPP)
• 272
Next Steps in the Evolution of GridICE: a Monitoring Tool for Grid Systems
GridICE is an open source distributed monitoring tool for Grid systems that is integrated in the gLite middleware and provides continuous monitoring of the EGEE infrastructure. The main goals of GridICE are: to provide both summary and detailed view of the status and availability of Grid resource, to highlight a number of pre-defined fault situations and to present usage information. In this paper, we briefly summarize the core ideas and design choices behind this tool. In particular, we describe the 3-layer data distribution architecture composed by intra-site and inter-site collection plus the end-user presentation; furthermore we list the number of supported measurements. After this introduction, we discuss the main focus of the paper, that is the description of the next steps in the GridICE development as driven by user requirements. The purpose is to disseminate our evolution strategy in order to solicit for a ommunity feedback.
Speaker: Dr Sergio Andreozzi (INFN-CNAF)
• 273
On using a generic framework for integrating advanced batch systems to production-level grid infrastructures
Advanced capabilities available in nowadays batch systems are fundamental for operators of high-performance computing centers in order to provide a high- quality service to their local users. Existing middleware allow sites to expose grid-enabled interfaces of the basic functionalities offered by the site’s computing service. However, they do not provide enough mechanisms for addressing the operational and scalability-related issues which are necessary for offering the same quality of service to the site’s grid users. This paper will present an overview of those issues and a proposed set of features required for the batch systems integration to a grid infrastructure. Then, we will describe a generic software framework that facilitates the implementation of those features. The framework is designed to interact with different grid execution services on one hand and with different batch systems on the other hand. Specialization is done via Java plug-ins, XSL plug-ins and XML configuration files. To illustrate the benefits of the framework features, an example of its usage for interfacing our site’s home-grown batch system (BQS) to the EGEE grid execution service (gLite CE) will also be presented. In particular, we will describe how the operational needs are addressed by the reusable components of the framework.
Speaker: Mr Sylvain Reynaud (IN2P3/CNRS)
• 274