Distributed computing enables you to implement scalable systems with a higher total computing power while it also increases the availability. As the components of a distributed system have to communicate via networks, it is important to understand the underlying protocols to use them efficiently. In this lecture we will go into details about (10) Gigabit Ethernet as this is still the most common LAN technology. In this context the most important Internet Protocols will be discussed and compared.
As nowadays networking performance is typically limited by the operating system, we will also discuss how Internet sockets are implemented in Linux and how alternatives based on Ethernet and other technologies look like.
Speaker:
Jonas Kunze(Johannes-Gutenberg-Universitaet Mainz (DE))
After a brief introduction to the history of web services, this lecture will cover the basics of REST (Representational State Transfer) and provide you with an understanding of essential terms and constraints. We will have a look at the API design process, and think about what you should consider when designing a scalable web service. Furthermore, we will explore how striving for a resource-oriented client architecture helps to reap the benefits of REST without sacrificing user experience.
Speaker:
Josef Hammer(CERN)
Slides
4:50 PM
→
5:50 PM
Building highly distributed systems within 5 minutes1h31/3-004 - IT Amphitheatre
Highly distributed systems are typically very complex. Traditionally, it took a long time to design the dataflow and it may have taken even more time to implement the necessary communication interfaces.
But using modern libraries to outsource the communication significantly reduces development and implementation time. During this lecture several communication patterns will be discussed and a selection of libraries for different application areas will be introduced:
Boost.Asio is a C++ library that provides you a consistent way to develop asynchronous communication and therefore makes it easy to develop a highly parallel program.
ØMQ is a library for many different programming languages. It provides the distribution of messages with several patterns and therefore clearly facilitates the development of distributed systems.
Apache Thrift is a framework enabling Remote Procedure Calls (RPCs) between many different languages. It generates source code for the server and client based on a given interface description file.
Speaker:
Jonas Kunze(Johannes-Gutenberg-Universitaet Mainz (DE))
The beauty, or bottom, quark is an extremely powerful object in high energy physics. Distinctive characteristics of the decay of the quark has motivated the design high-energy physics detectors in the quest to reconstruct the quark. Each component of the detector is used as a tool to identify these characteristics and present unique challenges that are solved with a combination of engineering, physics, and computing. These two lectures will explore the tools used and the different physics and computing environments they are used in, and the important interplay between the two environments and limitations they set on each will be considered. When all components are reassembled, the quark is described as a jet.
Speaker:
Tyler Mc Millan Dorland(Deutsches Elektronen-Synchrotron (DE))
This lecture gives a general overview on the evolution of detectors used in HEP and on the necessity of electronics in modern set-ups.
The second part of the lecture is focused on the Read-Out Electronics
(ROE) as a transmission line: information are transported from the detector to the Data Storage; at high frequency it is important to take into account particular characteristics (such as signal propagation and integrity, interaction between signals and electromagnetic
interferences) of the signal even if it is a digital one.
Speaker:
Francesco Messi
Slides
11:00 AM
→
11:30 AM
Coffee break
30m
31/3-009 - IT Amphitheatre Coffee Area
This lecture will begin by presenting the detecting systems needed for proper jet reconstruction. Then the jet reconstruction algorithms in detail are explained. A brief explanation of combination algorithms follows, completed by an examination of the datastructures used to store them. This will be followed by a brief description of a multivariate technique commonly used for the classification of these objects and the software package (TMVA) used to implement them. Particular attention is paid to which parts of these algorithms can be completed on university or personal level computing that is the common interface for analysts.
Speaker:
Tyler Mc Millan Dorland(Deutsches Elektronen-Synchrotron - DE)
This lecture will focus on electronics used in HEP: while the software tools available to program very performant firmware are more and more powerful, hardware limitations can not be (easily) overcome. High speed digital signals, today required in HEP setups, need to be handled as analog signals: signal propagation, interaction between signals and electromagnetic interferences must be carefully considered. Moreover, the power consuming of ASICs and FPGAs strongly depends of the operations required and may be a limitation of the board.
Speaker:
Francesco Messi
Slides
2:30 PM
→
3:30 PM
Introduction to machine learning and data mining1h31/3-004 - IT Amphitheatre
In this lecture, the basis of machine learning and data mining will be explained. Then, the most typical problems where machine learning is applied will be presented: classification, clustering, regression and anomaly detection. For those problems, some techniques that can be applied will be presented and briefly explained: decision trees, support machine vectors, k-NN, k-means and neural networks.
Speaker:
Juan Lopez Gonzalez(Universidad de Oviedo - ES)
In this lecture, the general concepts of self-organizing maps and its properties will be explained. Starting from the classic neural network approach, a MLP (introduced in the previous lecture), the concept of SOM will be explained. Its structure, the learning process and the later classification of the inputs for not seen cases. The main features of the maps: dimensional reduction and the conservation of the topological properties of the inputs, will be highlighted. Also, a small example will be shown where the attendants will see an actual map arranging itself and the resultant order will be interpreted. Finally, some other SOM based models will be shown to point out different architectures and possibilities.
Speaker:
Juan Lopez Gonzalez(Universidad de Oviedo - ES)