Grid5000:School2011 Program

See also: Spring School 2011 main page | Spring School 2011 program | Spring School 2011 practical sessions | registrations to practical sessions

Overview, with authors

This program is the final version. Practical session will be detailed on a registration site used to plan capacity. Initial list with more details is available.

To ease management of the event, please plan your arrival either

• between 13:30 and 14:00 on the Monday,
• between 10:30 and 11:00 or between 13:30 and 14:00 on the Tuesday.

• The green practical sessions are the sessions recommended to new users
• The orange practical sessions are contributed by users
• The first slot on the program each morning is for lessons.

	April 18th (Monday)	April 19th (Tuesday)				April 20th (Wednesday)				April 21st (Thursday)
9h-10h		Executive Committee meeting: April 6th review debriefing		Practical session: efficient customization of Grid'5000 : using Kadeploy3 on Grid'5000		Advanced tools for Grid'5000: taktuk Pierre Neyron				The future of Grid'5000 David Margery, Technical director
10h-10h30						Automating the measure of the NAS MG cache behavior on many architectures using CControl and Grid'5000 Swann Perarnau and Pierre Neyron.				Focus on networking: tools prepared by the technical staff
10h30-11h00		Registration/coffee break		coffee break
11h-11h30		Hemera working group: Methodology		Practical session: Grid'5000 is programmable (API Basics)		Experimental Study of a P2P B&B approach on top of Grid’5000 Mathieu Djamaï, Bilel Derbel and Nouredine Melab				Awards and Challenge winner's presentation
11h30-12h						Deploying a fault-tolerant computing middleware over Grid'5000: performance analysis of CONFIIT and its integration with a quantum molecular docking application Luiz Angelo Steffenel, Jean-Charles Boisson, Jaillet Christophe, Olivier Flauzac, Michael Krajecki, Chantal Barberot, Eric Henon and Stéphane Gérard.
12h00-12h30
12h30-13h30		Lunch break
13h30-14h	as from 13:30 : registration and coffee					Challenge: Leveraging BlobSeer to Boost up the Deployment and Execution of Hadoop Applications in Nimbus Cloud Environments on Grid'5000
14h-14h30	Introductory lessons	Hemera Scientific Challenges on Grid'5000
14h30-15h00						Challenge: Grid-in-a-grid: Deployment of a gLite Grid inside Grid'5000
15h00-15h30	Practical session: from ssh access to initial submissions	A Cost-Evaluation of MapReduce Applications in the Cloud Diana Moise, Alexandra Carpen-Amarie, Gabriel Antoniu and Luc Bouge.
15h30-16h00		Hybrid Distributed Computing Infrastructure Experiments in Grid5000: Supporting QoS in Desktop Grids with Cloud Resources Simon Delamare and Gilles Fedak.				Challenge: Deploy a Highly-dynamic Virtual Cluster Based on OpenNebula and Xen in Grid'5000
16h-16h30	Coffee Break
16h30-17h	Practical session: from ssh access to initial submissions	Put some Green in your Grid’5000 experiments !, Marcos Dias de Assuncao, Mohammed El Mehdi Diouri, Laurent Lefèvre, Olivier Mornard, Anne-Cécile Orgerie, Ghislain Landry, Tsafack Chetsa	Deploying and Using IaaS Clouds on Grid'5000, Alexandra Carpen-Amarie, Houssem-Eddine Chihoub, Pierre Riteau	Deploy a virtualized environment	Run MPI on Grid'5000	Creating Customized Grid'5000 Environments with Chef, Pierre Riteau	Large-Scale Trace Visualization Analysis with Triva and Pajé -- the G5K case study --, Lucas Mello Schnorr, Arnaud Legrand	Network isolation on Grid'5000	Use the API to gather metrics
17h-17h30
17h30-18h
18h-18h30
18h30-18h45
18h45-19h		Champagne cellar visit and tasting
19h00-20h						Official cocktail at City Hall
20h-...						Gala dinner

List of user presentations

• Experimental Study of a P2P B&B approach on top of Grid’5000. Mathieu Djamaï, Bilel Derbel and Nouredine Melab PDF
  • Abstract:: The state-of-the-art large scale approach for solving NP-hard permutation-like problems using parallel Branch-and-Bound (B&B) techniques are based on a Master-Slave model which is known to be limited in terms of scalability. To get beyond this limitation, we designed a Pure P2P approach which provides fully distributed mechanisms dealing with the B\&B algorithm.
    In this paper, we propose an experimental study of our approach to analyze its performances in both small-scale and large-scale environments. We compare our approach to a state-of-the-art Master-Slave approach and we study its intrinsic performances (Parallel Speed-up). We also study the impact of the network topology on the performances of our approach. The evaluation was carried out on the Aladdin/Grid’5000 testbed. Experimental results show an improved scalability, a reduced communication overhead and a satisfactory speed-up for our approach.
• A Cost-Evaluation of MapReduce Applications in the Cloud Diana Moise, Alexandra Carpen-Amarie, Gabriel Antoniu and Luc Bouge. PDF
  • Abstract:: MapReduce is a powerful paradigm that enables rapid implementation of a wide range of distributed data-intensive applications. Its open source implementation, the Hadoop project, has been recently adopted by the Cloud computing community. This paper focuses on evaluating the cost of moving MapReduce applications to the Cloud, in order to find a proper trade-off between cost and performance for this class of applications. We provide a cost evaluation of running MapReduce applications in the Cloud, by looking into two aspects: the overhead implied by executing the job on the Cloud, compared to executing it on a Grid, and the actual costs of renting cloud resources. We compared the runtime of two MapReduce applications when they are executed with the Hadoop framework, in two similar environments: clusters belonging to the Grid'5000 platform and in a Nimbus Cloud deployed on top of Grid'5000 nodes.
• Hybrid Distributed Computing Infrastructure Experiments in Grid'5000: Supporting QoS in Desktop Grids with Cloud Resources Simon Delamare and Gilles Fedak. PDF
  • Abstract:: Hybrid Distributed Computing Infrastructures (DCIs) allow users to combine Grids, Desktop Grids, Clouds, etc. to obtain for their users even larger computing capabilities. In this paper, we present an experimental study of the SpeQuloS framework which aims at providing QoS to Desktop Grid by provisioning on-demand Cloud resources. We describe the experimental platform which relies on Grid5000 to mimic both a Desktop Grid system and a Cloud system. Preliminary results are presented which shows the potential of the SpeQuloS approach.
• Deploying a fault-tolerant computing middleware over Grid'5000: performance analysis of CONFIIT and its integration with a quantum molecular docking application Luiz Angelo Steffenel, Jean-Charles Boisson, Jaillet Christophe, Olivier Flauzac, Michael Krajecki, Chantal Barberot, Eric Henon and Stéphane Gérard. PDF
  • Abstract:: P2P computing middlewares are interesting options for grid computing applications that require scalability and resiliency. Nevertheless, most P2P computation systems rely on partially centralized or hybrid decentralized architectures to distribute tasks and collect the results, raising fault tolerance and bottleneck issues. CONFIIT (Computation Over Network with Finite number of Independent and Irregular Tasks) is a purely decentralized middleware for grid computing, relying on a virtual ring for topology management and for task scheduling. Firstly this paper evaluates the impact of node placement and task granularity on the performance of CONFIIT while solving the well-known Langford permutation problem. Secondly, an application of CONFIIT on a complex real life problem is presented.
• Automating the measure of the NAS MG cache behavior on many architectures using CControl and Grid'5000 Swann Perarnau and Pierre Neyron. PDF
  • Abstract: The NAS Parallel Benchmarks are a well recognized set of parallel benchmarks, implementing classical HPC algorithms. Understanding the memory behavior of such algorithms enables the development of better architectures and better parallel runtimes. In this paper we discuss the use of Grid'5000 for the various system architectures it features, to measure and understand the memory behavior of the NAS benchmarks. We achieve that thanks to a tool we recently developed: CControl, which allows to control the cache available to an application. Using this tool we present several measurements of the working sets of the NAS MG program, showing how its performance evolves when the cache conguration varies. While we would have liked to gather a wide range of results using the different architectures featured by the platform for comparison, we discovered that unfortunately such measurements cannot be fully automated on Grid'5000 yet. We explain why.