Program

Large-Scale Distributed Systems, from Clusters to Grids
Time	Description
09:00 - 10:00	High Performance Computing - Resource and Job Management for HPC
10:00 - 10:45	Grid'5000 presentation talk
10:50 - 13:00	Grid'5000 first practical session -- Basics
13:00 - 14:00	Lunch
14:30 - 17:00	Grid'5000 second practical session -- Virtualization
17:15 - 18:00	Experimentation problematics (on Grid'5000 but not only)
19:00 - 20:00	Dinner

High Performance Computing - Resource and Job Management for HPC

09:00 - 10:00
Speaker: Yiannis Georgiou

This presentation makes an introduction on the basic concepts of High Performance Computing platforms - Cluster Computing, Grids and Clouds by analyzing characteristics and differences between them. Then we enter into detail upon the particular middleware called Resource and Job Management System which has a strategic position on the HPC stack and we analyze the OAR system.

Grid'5000 presentation talk

10:00 - 10:45
Speaker: Joseph Emeras & Yiannis Georgiou

During this talk we will present the Grid'5000 platform that will be used during the day for the practical sessions. The goal is to give a quick overview of how it works, the different features, its organization and how it should be used.

Grid'5000 first practical session -- Basics

10:50 - 13:00
Speaker: Joseph Emeras & Yiannis Georgiou

During this practical session, the students will learn the basics of Grid'5000: how to configure ssh for an easier use, how to submit jobs, deploy some default environments and customize their own environments and make them available for future use.
Links to the tutorials:
https://www.grid5000.fr/mediawiki/index.php/SSH_quick_configuration
http://www.grid5000.fr/mediawiki/index.php/Cluster_experiment-OAR2
https://www.grid5000.fr/mediawiki/index.php/Deploy_environment-OAR2

Grid'5000 second practical session -- Virtualization

14:30 - 17:00
Speaker: Joseph Emeras

This session will present different ways to virtualize an environment on Grid'5000: the Xen and KVM tools will be presented. The goal of this session is to give to the students an overview of the problematics encountered when virtualizing in large scale platforms. It proposes several solutions to deal with problems such as managing the network for the virtual machines. A disk space consideration problematic is also presented when multiplicating the virtual machines on a single node and a solution based on LVM and its snapshotting system is presented for the Xen hypervisor.
Links to the tutorials:
https://www.grid5000.fr/mediawiki/index.php/Deploy_a_virtualized_environment
https://www.grid5000.fr/mediawiki/index.php/Deploy_a_virtualized_environment_--_other_alternatives

Experimentation problematics (on Grid'5000 but not only)

17:15 - 18:00
Speaker: Joseph Emeras

This talk presents an overview of the experimentation problematics such as experiment result reproducibility, environment reconstructability and gives some advices to get valuable results in experimenting on a platform such as Grid'5000. It then states several of the tricks and treats that student will face when using Grid'5000 and gives some useful tips for improving their experiment like using vlans for isolating their network to prevent from being disturbed in their experiment by the regular network traffic.