CERN School of Computing 2014

24 Aug 2014, 14:00 → 6 Sept 2014, 02:40 Europe/Lisbon

University of Minho and LIP

Alberto Pace (CERN)

List of participants 9a-List_of_participants.pdf

Participants Bios A-F Participants_Bio_Pictures_A_F.pdf

Participants Bios G-M Participants_Bio_Pictures_G_M.pdf

Participants Bios N-S Participants_Bio_Pictures_N_S.pdf

Participants Bios T-W Participants_Bio_Pictures_T_W.pdf

Sunday 24 August

14:00 Arrival

20:45 Dinner at Hotel

Monday 25 August

08:15 Bus departure from Hotel Melia to Braga

08:40 Visit of the exhibition

60 years of CERN and Portuguese membership

Convener: Alberto Pace (CERN)

1 Intervention of the Rector of University of Minho

Speaker: Antonio Cunha

2 Intervention of the President of LIP (Laboratorio Experimental de Fisica de Particulas)

Speaker: Jose Mariano Gago

3 Intervention of the Portuguese National Science Foundation, FCT (on behalf of the Ministry of Education and Science),

Speaker: Miguel Seabra

4 CERN, 60 years of Science for Peace

Speaker: Rolf Heuer (CERN General Director)

10:30 Coffee break

60 years of CERN and Portuguese membership

Convener: Antonio Pina

5 Challenges of the Computing Infrastructures for Scientific Research of the Coming Years

Speaker: Frederic Hemmer (CERN, IT Department Head)

6 The Cern School of Computing

Speaker: Alberto Pace (CERN, Director of CSC)

7 Introduction to Physics computing

Speaker: Arnulf Quadt (Georg-August-Universitaet Goettingen (DE))

12:30 Bus back to University of Minho

12:50 Lunch

Data Analysis

8 Tools and Techniques - Lecture 1

Introduction to the Track; Tools You Can Use We discuss some of the characteristics of software projects for high energy physics, and some of the issues that arise when people want to contribute to them. This forms the framework for the Software Technologies Track. We then discuss several categories of tools & techniques you can use to make yourself more productive and effective. Continuous testing and documentation has proven to be important in producing high quality work, but it's often difficult to do; we discuss some available approaches. Many problems require specific tools and techniques to solve them effectively: We discuss the examples of performance tuning and memory access problems. Throughout, we discuss software engineering from the perspective of the individual contributor, both as a formal process and how it actually effects what you do.

Speaker: Robert Jacobsen (Lawrence Berkeley National Laboratory (LBNL))

9 Tools and Techniques - Lecture 2

Tools for Collaboration HEP software is built by huge teams. How can this be done effectively, while still giving people satisfying tasks to perform? This lecture discusses some of the technical approaches used. Source control (e.g. SVN, Git) tools are becoming common. We discuss the different ways they can be used with their advantages and disadvantages. We then address the larger area of release control techniques (e.g. CMT, CMake) and release testing & distribution. Our focus on why is this considered a hard problem, and what the current issues are when dealing with it. We close with a summary of observations.

Speaker: Robert Jacobsen (Lawrence Berkeley National Laboratory (LBNL))

10 Presentation of the School

Speaker: Alberto Pace (CERN)

16:20 Drinks and fruits

Data Analysis

11 Tools and Techniques - Exercise 1

The first exercises provide some direct experience with the individual tools and techniques described in Lectures 1 and 2. Teams of two students will work together to update existing applications, working through examples designed to show the strengths and weaknesses of various tools and approaches. This will be followed by small projects for additional development experience. Exercise 3 starts work with code management and release tools.

Speaker: Robert Jacobsen (Lawrence Berkeley National Laboratory (LBNL))

12 Tools and Techniques - Exercise 2

The first exercises provide some direct experience with the individual tools and techniques described in Lectures 1 and 2. Teams of two students will work together to update existing applications, working through examples designed to show the strengths and weaknesses of various tools and approaches. This will be followed by small projects for additional development experience. Exercise 3 starts work with code management and release tools.

Speaker: Robert Jacobsen (Lawrence Berkeley National Laboratory (LBNL))

18:45 Walk back to the hotel

19:30 Bus from Melia Braga Hotel to Bom Jesus

19:45 Welcome Cocktail

20:45 Dinner

23:00 Bus back to Melia Hotel

Tuesday 26 August

Data Analysis

13 Introduction to Physics Computing - Lecture 2

Speaker: Arnulf Quadt (Georg-August-Universitaet Goettingen (DE))

Base Technologies

14 Computer Architecture & Perf. Tuning - Lecture 1

Understanding scalable hardware The lecture describes the hardware architecture of a modern x86_64 PC server. Architectures from other companies, such as Nvidia and ARM, will also be mentioned. Acceleration opportunities (but also bottlenecks) in the architecture will be covered in detail with an aim to give the students a good understanding of what resources are available from a hardware viewpoint. We will also discuss several strategies which can allow software to scale to the maximum resource potential in a given architecture. These strategies are based on both data and task parallelism. We will stress the importance of a Data Oriented Design and also mention the issue of “performance portability” across platforms. Some important factors related to programming styles will be reviewed. To back up everything with evidence, a couple of scalable examples from physics will be portrayed.

Speaker: Mr Sverre Jarp (CERN)

10:45 Updates

11:00 Coffee

Base Technologies

15 Computer Architecture & Perf. Tuning - Lecture 2

"Architectural Details and Performance Studies" Considering the rise of complex many-core processors, a sufficient understanding of their architecture and of the relevant performance tuning opportunities has become an indispensable element of software development. Although by using various tools we are often able to get a generous peek both inside the hardware and software, drawing high-level conclusions that impact our software is not always straightforward. Another considerable challenge comes from accelerators and co-processors, where different computing paradigms take precedence, such as extreme data parallelism, different math capabilities, the importance of special languages as well as memory size and topology changes. The objective of this lecture is to bring the audience closer to "where it matters" in modern computing systems, and to identify important paths for high performance.

Speaker: Andrzej Nowak (CERN)

12:30 Lunch

13:30 Study time* and/or Daily Sports Program

Optional session for self-studying, during which lecturers may be available on-demand. Interested students should make an explicit appointment with the lecturer at least the day before.

15:45 Drinks and fruits

Data Analysis

16 Tools and Techniques - Exercise 3

The first exercises provide some direct experience with the individual tools and techniques described in Lectures 1 and 2. Teams of two students will work together to update existing applications, working through examples designed to show the strengths and weaknesses of various tools and approaches. This will be followed by small projects for additional development experience. Exercise 3 starts work with code management and release tools.

Speaker: Robert Jacobsen (Lawrence Berkeley National Laboratory (LBNL))

17 Tools and Techniques - Exercise 4

After the two-person teams acquire additional experience with the code management and release tools in exercise 4 and 5, in exercise 6 we will group projects to demonstrate some of the real-world issues discussed in the lecture. Groups of two teams will first work together to create a functional release from individual sub-projects at various stages of completion to show the strengths and weaknesses of test and release tools. This is followed by a larger scale exercise with groups of multiple teams.

Speaker: Robert Jacobsen (Lawrence Berkeley National Laboratory (LBNL))

18 Tools and Techniques - Exercise 5

After the two-person teams acquire additional experience with the code management and release tools in exercise 4 and 5, in exercise 6 we will group projects to demonstrate some of the real-world issues discussed in the lecture. Groups of two teams will first work together to create a functional release from individual sub-projects at various stages of completion to show the strengths and weaknesses of test and release tools. This is followed by a larger scale exercise with groups of multiple teams.

Speaker: Robert Jacobsen (Lawrence Berkeley National Laboratory (LBNL))

19 Tools and Techniques - Exercise 6

After the two-person teams acquire additional experience with the code management and release tools in exercise 4 and 5, in exercise 6 we will group projects to demonstrate some of the real-world issues discussed in the lecture. Groups of two teams will first work together to create a functional release from individual sub-projects at various stages of completion to show the strengths and weaknesses of test and release tools. This is followed by a larger scale exercise with groups of multiple teams.

Speaker: Robert Jacobsen (Lawrence Berkeley National Laboratory (LBNL))

20:00 Free time

20:45 Dinner at hotel

Wednesday 27 August

Base Technologies

20 Software Design in the Many-Cores era - Lecture 1

Outfitting a Modern HEP Data Processing Framework for Concurrency Even though the miniaturization of transistors on chips continues like predicted by Moore's law, computer hardware starts to face scaling issues, so-called performance 'walls'. Probably, the best known is the 'power wall', which limits clock frequencies. Amongst others, a way of increasing processor performance remains now to integrate many cores in the same chip. At the same time, the upcoming LHC upgrade will increase the required CPU power drastically. Both problems challenge the current way of software design in high energy physics (HEP). Developers in high energy physics are forced to re-think their ways of software design and need to move to massively parallel applications. This lecture will explain the current HEP software design, the hardware and physics issues that need to be tackled, and possible approaches to achieve the required level of parallelization.

Speakers: Benedikt Hegner (CERN), Danilo Piparo (CERN)

21 Secure Software - Lecture 1

Speaker: Sebastian Lopienski (CERN)

10:45 Updates

11:00 Coffee

Base Technologies

22 Secure Software - Lecture 2

Speaker: Sebastian Lopienski (CERN)

12:30 Lunch

13:30 Study Time* and/or Daily Sports Program

Optional session for self-studying, during which lecturers may be available on-demand. Interested students should make an explicit appointment with the lecturer at least the day before.

16:00 Drinks and fruits

Base Technologies

23 Secure Software - Exercise 1

Speakers: Dr Giuseppe Lo Presti (CERN), Sebastian Lopienski (CERN)

24 Secure Software - Exercise 2

Speakers: Dr Giuseppe Lo Presti (CERN), Sebastian Lopienski (CERN)

25 Secure Software - Exercise 3

Speakers: Dr Giuseppe Lo Presti (CERN), Sebastian Lopienski (CERN)

19:45 Free time

20:45 Dinner at hotel

Thursday 28 August

Base Technologies

26 Software Design in the Many-Cores era - Lecture 2

Base Concepts of Parallel Programming: A Pragmatic Approach This lecture will explain the main concepts behind concurrent programming. First, a theoretical introduction into threads will be given. As the new C++ standard (C++11) now provides built-in support for parallel programming, the new features of this standard will be shown. The second part of this lecture builds thread-safety and concrete technologies to tackle the problem of concurrent data access.

Speakers: Benedikt Hegner (CERN), Danilo Piparo (CERN)

27 Software Design in the Many-Cores era - Lecture 3

Development Patterns for Parallel Software Development The focus of this lecture lies on repeating a few design patterns of sequential software. It then discusses under which conditions these can be transformed into parallel design patterns. It discusses how different level of constrains affect the scaling of the parallel patterns shown.

Speakers: Benedikt Hegner (CERN), Danilo Piparo (CERN)

10:45 Updates

11:00 Coffee

Base Technologies

28 Secure Software - Lecture 3

Web App. Security Debriefing

Speaker: Sebastian Lopienski (CERN)

12:30 Bus to Povoa do Lanhoso (sport afternoon)

13:30 Sports afternoon

Details of Sports options,
schedule, name of participants TBC

19:00 Special outside dinner

Friday 29 August

Base Technologies

29 Software Design in the Many-Cores era - Lecture 4

Understanding, Debugging and Profiling a Complex Multithreaded Application Writing thread-safe code is a complex problem and difficult to master. This lecture explains basic tools and techniques assisting you in parallel software development. Firstly, we will show basic examples and tools for static code analysis. Then we will have a look at how to understand and debug a multithreaded application with th GNU debugger. Finally, we will tackle performance measurements of applications and how to identify which parts deserve speed improvement or parallelization.

Speakers: Benedikt Hegner (CERN), Danilo Piparo (CERN)

30 Software Design in Many-core area - Exercise 2

Speakers: Benedikt Hegner (CERN), Danilo Piparo (CERN)

10:45 Updates

11:00 Coffee

Base Technologies

31 Software Design in Many-core area - Exercise 3

Speakers: Benedikt Hegner (CERN), Danilo Piparo (CERN)

12:30 Lunch

13:30 Study Time* and/or Daily Sports Program

Optional session for self-studying, during which lecturers may be available on-demand. Interested students should make an explicit appointment with the lecturer at least the day before.

16:00 Drinks and fruits

Base Technologies

32 Software Design in Many-core area - Exercise 3

Speakers: Benedikt Hegner (DESY), Danilo Piparo (CERN)

33 Computer Architecture & Perf. Tuning - Exercise 1

Speakers: Andrzej Nowak (CERN), Mr Sverre Jarp (CERN)

34 Computer Architecture & Perf. Tuning - Exercise 2

Speakers: Andrzej Nowak (CERN), Mr Sverre Jarp (CERN)

19:45 Free time

20:45 Dinner at hotel

Saturday 30 August

Data Analysis

35 Data Analysis - Lecture 1

Speaker: Ivica Puljak (University of Split (HR))

36 Data Analysis - Lecture 2

Speaker: Ivica Puljak (Technical University of Split FESB)

10:45 Updates

11:00 Coffee

Data Analysis

37 Data Analysis - Lecture 3

Speaker: Ivica Puljak (Technical University of Split FESB)

12:30 Lunch

13:30 Free Time

20:45 Dinner at hotel

Sunday 31 August

08:00 Full Day Excursion

Programme and schedule TBC

20:00 Special Outside Dinner

Monday 1 September

Data Technologies

38 Data Technologies - Lecture 1

Speaker: Alberto Pace (CERN)

39 Data Technologies - Lecture 2

Speaker: Alberto Pace (CERN)

10:45 Updates

11:00 Coffee

Data Analysis

40 Data Analysis - Lecture 4

Speaker: Ivica Puljak (Technical University of Split FESB)

12:30 Lunch

13:30 Study Time* and/or Daily Sports Program

Optional session for self-studying, during which lecturers may be available on-demand. Interested students should make an explicit appointment with the lecturer at least the day before.

16:00 Drinks and fruits

Data Analysis

41 Data Analysis - Exercise 1

Speaker: Ivica Puljak (Technical University of Split FESB)

42 Data Analysis - Exercise 2

Speaker: Ivica Puljak (Technical University of Split FESB)

43 Data Analysis - Exercise 3

Speaker: Ivica Puljak (Technical University of Split FESB)

19:45 Free time

20:45 Dinner at hotel

Tuesday 2 September

Data Technologies

44 Data Technologies - Lecture 3

Speaker: Alberto Pace (CERN)

45 Data Technologies - Lecture 4

Speaker: Alberto Pace (CERN)

10:45 Updates

11:00 Group photo

11:15 Coffee

Data Technologies

46 Spotlight on Big Data/Technologies -... inspired by nature, society, physics and mathematics

A look behind Big Data and Cloud technology and into the future of money and civilization archiving tools.

Speaker: Mr Andreas Joachim Peters (CERN)

12:30 Lunch

13:30 Study time and/or Daily Sports Program

Optional session for self-studying, during which lecturers may be available on-demand. Interested students should make an explicit appointment with the lecturer at least the day before.

16:00 Drinks and fruits

Data Technologies

47 Data Technologies - Exercise 1

Speakers: Alberto Pace (CERN), Mr Andreas Joachim Peters (CERN)

48 Data Technologies - Exercise 2

Speakers: Alberto Pace (CERN), Mr Andreas Joachim Peters (CERN)

49 Data Technologies - Exercise 3

Speakers: Alberto Pace (CERN), Mr Andreas Joachim Peters (CERN)

19:45 Free time

20:45 Dinner at hotel

Wednesday 3 September

Data Analysis

50 Data Analysis - Lecture 5

Speaker: Ivica Puljak (Technical University of Split FESB)

51 Data Analysis - Lecture 6

Speaker: Ivica Puljak (Technical University of Split FESB)

10:45 Updates

11:00 Coffee

Data Technologies

52 Data Technologies - Lecture 5

Speaker: Alberto Pace (CERN)

12:30 Lunch

Data Analysis

53 Multivariate Visualisation - Lecture 1

Speaker: Benjamin Radburn-Smith (Purdue University (US))

14:30 Study time* and/or Daily Sports Program

Optional session for self-studying, during which lecturers may be available on-demand. Interested students should make an explicit appointment with the lecturer at least the day before.

19:45 Free time

20:45 Dinner at hotel

Thursday 4 September

Data Technologies

54 Data Technologies - Exercise 4

Speakers: Alberto Pace (CERN), Mr Andreas Joachim Peters (CERN)

55 Data Technologies - Exercise 5

Speakers: Alberto Pace (CERN), Mr Andreas Joachim Peters (CERN)

10:45 Updates

11:00 Coffee

Data Analysis

56 Data Analysis - Exercise 4

Speaker: Ivica Puljak (Technical University of Split FESB)

12:30 Lunch

Data Analysis

57 Data Analysis - Exercise 5

Speaker: Ivica Puljak (Technical University of Split FESB)

14:30 Free time

16:00 Drinks and fruits

16:30 Optional Examination

17:45 Bus departure for the football match

18:15 Traditional CSC Football Match

19:45 Bus to Hotel Melia

20:00 Free time

20:45 Dinner

Friday 5 September

Presentations by students

58 Calculability and comlexity

Brief introduction of the analisys of calculability and complexity, and introduction of theoretical computer science. Turing machine, complexity hierarchy, etc...

Speaker: Stefano Martina (Universita e INFN (IT))

59 Challenges in the CERN Accelerator Complex

Beam instabilities and loss of Landau damping are major limiting factors for the HL-LHC era. A short introduction to how the beam is produced and what the major challenges are.

Speaker: Helga Timko (CERN)

60 Clustering

In the era before 'Big data' LHC physicists used clustering algorithms to join 3 dimensional objects together inside their detectors. Now similar algorithms, working in N dimensions, are revolutionising the way that people choose which TV to watch, how the business world makes transactions and even how geeks find love. Clustering can be used to create structures that isolate nuances in the larger data set, infer the properties of missing variables, or split the data into smaller chunks for which more focused algorithms can be optimised. The science of clustering algorithms focus on how to both how such applications can be optimised, how clusters can be recombined combined and how to verify the validity of the properties that these clusters suggest.

Speaker: Vincent Alexander Croft (Radboud University Nijmegen (NL))

61 Modern HEP Data Analysis - How to handle the growing complexity

In the era before 'Big data' LHC physicists used clustering algorithms to join 3 dimensional objects together inside their detectors. Now similar algorithms, working in N dimensions, are revolutionising the way that people choose which TV to watch, how the business world makes transactions and even how geeks find love. Clustering can be used to create structures that isolate nuances in the larger data set, infer the properties of missing variables, or split the data into smaller chunks for which more focused algorithms can be optimised. The science of clustering algorithms focus on how to both how such applications can be optimised, how clusters can be recombined combined and how to verify the validity of the properties that these clusters suggest.

Speaker: Raphael Marius Friese (KIT - Karlsruhe Institute of Technology (DE))

62 Non-parametric density estimation

Fitting is often used to model data distributions of different categories in order to identify, or unfold, these components in regions of the parameter space where they are mixed. The traditional use of high-order polynomial functions is now being replaced by non-parametric techniques as the Kernel Density Estimation (KDE) and Density Estimation Trees (DETs). Among the advantages of these technique there is the applicability in multi-dimensional, multi-modal datasets. I briefly introduce the concept of non-parametric density estimation, and of the KDE and DET techniques taking examples of applications from recent HEP papers.

Speaker: Lucio Anderlini (Laboratoire d'Annecy de Physique de Particules)

63 Parallelism for free with Haskell

Haskell is a standardised, general-purpose purely functional programming language with non-strict semantics and strong static typing. In this talk some of these properties will be demonstrated by working through a very simple example. Having these properties, it is possible to build safe and highly parallel applications running on multi-core architectures or GPUs without significant effort from the developer's side. This concept will also be illustrated by slightly changing the original example to magically turn it into a parallel application. The talk will be summed up by some useful links, references and information regarding learning material, upcoming events and how to become involved in the Haskell community.

Speaker: Janos Daniel Pek (CERN)

11:00 Coffee break

Presentations by students

64 Performance of RAID arrays

1) interfaces 2) low productivity RAID 5 3) SSD 4) Inefficiency of the classic RAID with SSD

Speaker: Michael Sokolov (SRV24)

65 Profiling code in Python

Python code is much easier to write than C, yet much less efficient. It's often assumed that Python is not performance-oriented and therefore making effort to optimize it doesn't pay off. However, if there arises a need to profile the code and find bottlenecks we are not completely lost. In this talk I will tell why perf(_deluxe.py) is not suitable for this purpose and which tools can we take instead

Speaker: Pawel Szostek (CERN)

66 Puppet is so awesome it sucks

Puppet is IT automation software that defines and enforces the state of your infrastructure throughout your software development cycle. From provisioning and configuration to orchestration and reporting, from initial code development through production release and updates, Puppet frees sysadmins from writing one-off, fragile scripts and other manual tasks. At the same time, Puppet ensures consistency and dependability across your infrastructure. With Puppet, repetitive tasks are automated away, so sysadmins can quickly deploy business applications, scaling easily from tens of servers to thousands, both on-premise and in the cloud. In the talk we will see why this software's idea is so awesome.

Speaker: Hristo Mohamed (University of Sofia (BG))

67 Software Defined Networking - programming your network

Software Defined Networking is a new model for managing and controlling networks that emerged from the need to make the network more agile in virtualized datacentre environments. Given the short time of the presentation I will present the concept of SDN (and OpenFlow, the main underlying protocol), by analogy to traditional networking, highlighting the main advantages (centralized control, hence full/better control of the network), yet point out potential caveats (scalability issues, limitations in hardware deployments).

Speaker: Stefan Nicolae Stancu (CERN)

68 Toward construction of neutron diffractometer for protein crystal

We performed a time-of-flight (TOF) single crystal neutron diffraction experiment with a diffractometer (the IBARAKI Biological Crystal Diffractometer (iBIX)) installed at a coupled moderator (CM) pulsed neutron source in J-PARC using single crystal silicon, and we determined several candidates for fundamental fitting functions to faithfully reproduce the TOF Bragg reflection profile asymmetries with the longer tail shape. The Vavilov and Landau distributions used to describe the energy loss of charged particles traversing a thin absorber were found to be in excellent agreement with the observed TOF profile. We are planning to design a new TOF single crystal diffractometer installed at a decoupled moderator (DM) pulsed neutron source in J-PARC. The peak profile provides narrow neutron pulses with short tail. In any event, however, it is expected that the Vavilov and Landau functions are very effective and appropriate for use in the TOF distribution of Bragg reflections because there is no fundamental difference in the neutron moderation process between the two kinds of moderator. It is possible to make use of this functions for the integration of Bragg reflection in the case of profile-fitting refinement for protein sample; this functions might be also be applicable to peak separation of the overlapped Bragg reflection in the TOF direction in foreseeable future.

Speaker: Tomoyori Katsuaki (KEK)

12:30 Lunch

14:00 Special Session with Hosting University Dignitaries

Vice Rector of University of Minho: Professor Rui Vieira de Castro and
Professor Gaspar Barreira, LIP

14:30 Closing Ceremony

16:00 Drinks and fruits

16:30 Free time

19:30 Departure to closing dinner

19:50 Closing Reception and Dinner