Grid'5000 user report for Raphael Couturier

Jump to: navigation, search

User information

Raphael Couturier (users, user, account-manager, lyon, ml-users user)
More user information in the user management interface.


  • solving PDE problem with asynchronous iterations (Application) [in progress]
    Description: This challenge consists in using iterative methods to solve PDE problem. Using distant clusters which may be heterogeneous prevents from using frequent synchronizations. In order to let machines computes freely their iterations one can use asynchronous iteration model. With the CRAC library developped by S. Domas (LIFC - Belfort), it is possible to develop PDE solver using the Multisplitting Newton method that only require one synchronization at each time step. Currently, the asynchronous version is in average twice fast as the synchronous one with more than 100 cpu using 2 or 3 sites.
  • GREMLINS - GRid Efficient Methods for LINear Systems (Application) [in progress]
    Description: Sparse linear systems often appear in numerous scientific applications. Grid computing is an answer to the growing demand of computational power in many scientific domains (mechanic, biology, ...) in order to solve such large linear systems. Unfortunately, the heterogeneity of the machines and the variability of the interconnection networks bring new algorithmic problems. The goal of this project is to define new algorithmic schemes to solve sparse linear systems on heterogeneous and distributed clusters. Those algorithms will be implemented in a library which will be freely available for the scientific community. To achieve this goal, the multisplitting method which consists in decomposing the linear system into several sub-systems will be used. In this method, the resolution takes an iterative form by applying on each processor a sequential method (direct or iterative) to solve its sub-system until the global result becomes stable. This method can be used either in synchronous or in asynchronous mode. In the latter mode, processors work independently and use the last received data from their neighbours in their computations. However, this method is only applicable to some matrices with a particular spectral radius. To avoid this restriction, we aim at studying the influence of pre-processing techniques such as renumbering and pre-conditioning.
    More information here
  • Java and asynchronous iterative applications: large scale experiments (Application) [in progress]
    Description: This work focuses on large scale experiments with Java and asynchronous iterative applications. In those applications, tasks are dependent and the use of distant clusters may be difficult, for example, because of latencies, heterogeneity, and synchronizations. Experiments have been conducted on the Grid'5000 platform using a new version of the Jace environment. We study the behavior of an application (the Poisson problem) with the following experimentation conditions: one and several sites, large number of processors (from 80 to 500), different communication protocols (RMI, sockets and NIO), synchronous and asynchronous model. The results we obtained, demonstrate both the scalability of the Jace environment and its ability to support wide-area deployments and the robustness of asynchronous iterative algorithms in a large scale context.
    Results: In this work we have described a set of large scale experiments conducted on asynchronous iterative applications and with a Java-based execution environment. We show how this kind of algorithms are well suited for large scale grid contexts compared to synchronous ones. We also show how the new version of the Jace platform successfully faces the scalability issues. Our current work focuses on developing our experiments and analysis. We are testing Jace V2 with other scientific applications and we are refining our quantitative analysis by measuring different parameters such as bootstrapping time, synchronizations time.
    More information here
  • Java asynchronous computation environment (Middleware) [in progress]
    Description: A Comparative Study between our environment Jace and MPJ Express to solve scientific applications.
  • Evaluation of Mapping algorithms for JaceP2P (Middleware) [achieved]
    Description: The aim of this experimentation is to evaluate the performance of several mapping algorithms in the JaceP2P middleware on compare them. Algorithms evaluated are : a simple one (basic), a modified version of QM (Quick Quality Map) -- adapted to our model -- and an algorithm based on the edgecuts principles. The application is based on the asynchronous iterations model.
    Results: First results : gains (in execution time) provided by each mapping algorithm compared to the no mapping version of JaceP2P : simple : 20% QMm : 45%



    Success stories and benefits from Grid'5000

    • Overall benefits
    • Grid5000 offers an important number of distants sites that enables to point out the advantages of asynchronous iterative methods. Using distant clusters with problems for which dependencies are not negligible is often difficult.

    last update: 2010-08-20 08:32:52