# Grid'5000 user report for

## User information

( user)*More user information in the user management interface.*

## Experiments

**GREMLINS (Networking) [in progress]**

*Description:*Numerical applications in different scientific domains (biology mechanics, geophysics, ..) need a more and more increasing computational power in order to simulate phenomena close to the reality. They use more often existing sparse linear system librairies and tools which are efficient on sequential machines, on parallel computers or on PC-clusters. There exists in labs unused local computational power via PC's or PC clusters which can be federated into into an heteregenous and distributed grid. The grid computing seems to be a cheap answer of the computational power demand. 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. The communications being penalizing on such distributed clusters, the new algorithms will have to be coarse grained in order to minimize the former. 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 reordering,load balancing and pre-conditioning methods.*Results:***Migrating a large-scale numerical application to the Grid5000 environment (Application) [in progress]**

*Description:*The primary aim of this project is to investigate how a large scale atomic physics application can be migrated to a grid environment, such as Grid5000, to provide an effective and numerically robust tool to compute accurate atomic data. A novel aspect of this work is the rigorous investigation of the numerical validation of large scale distributed scientific computation. This is a topic of considerable importance, but one that has received relatively little attention in the computational science community. The project brings together tools and expertise from two international research groups. LIP6 will provide the CADNA library, a software tool based on stochastic discrete arithmetic that is used to validate the stability of numerical software. Queen's University will provide 2DRMP, a suite of software based on R-matrix techniques that is used to simulate electron collisions with hydrogen-like atoms and ions at intermediate energies. Preliminary work has already been undertaken. The CADNA library has been integrated into a selection of subroutines in sequential parts of the 2DRMP package to investigate numerical instability. In particular, the algorithm to compute two dimensional integrals has been modified to remove the numerical instabilities identified by CADNA. Work is in progress to implement CADNA within the complete package. Within this project we propose to employ MPI CADNA, a variant of CADNA that has been used to study the robustness of numerical codes running on parallel computers. In particular, we wish to use the MPI CADNA library to develop a robust numerical code in the grid computing context. The main features of 2DRMP are described in the following sections, in particular those computational aspects that make 2DRMP an attractive application to deploy in a large grid environment. 2DRMP has the potential to use many processors. For example, a typical computation could involve 200 simultaneous matrix digonalisations. If 4 and 8 processors were devoted to each diagonalisation this would occupy 800 and 1600 processors respectively.*Results:*

## Publications

**e-Collisions Using e-Science [to a]***(international)**EntryType:*inproceedings *Author:*N.S Scott, V. Faro-Maza, M.P. Scott, T. Harmer, J-M. Chesneaux, C. Denis, F. Jezequel *Booktitle:*Mathematical Modeling and Computational Physics *Address:*High Tatra Mountains, Slovakia **Efficiently solving large sparse linear systems on a distributed and heterogeneous grid by using the multisplitting-direct method [to a]***(international)**EntryType:*inproceedings *Author:*Contassot-Vivier, S. and Couturier, R. and Denis, C. and Jézéquel, F *Booktitle:*in Proc. 4th International Workshop on Parallel Matrix Algorithms and Applications *Address:*Rennes, France **A sparse linear system solver used in a distributed and heterogeneous grid computing environment [2008]***(international)**EntryType:*inproceedings *Equipe:*and *Author:*Denis, Christophe and Couturier, Raphaël and Jézéquel, Fabienne *Booktitle:*INSYS'08 *Pages:****--*** **GREMLINS: a large sparse linear solver for grid environment [2008]***(international)**EntryType:*article *Equipe:*and *Author:*Couturier, R. and Denis, C. and Jézéquel, F. *Journal:*Parallel Computing *Pages:*380--391 *Volume:*34 *Number:*6-8

## Collaborations

## 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: 2006-08-30 15:48:45*