Skip to Main content Skip to Navigation
Theses

Couplage à hautes performances de codes parallèles et distribués

Abstract : The quick power's growth of current computers and their interconnection in clusters and computational grids using high speed networks, allow to simulate more complex physical phenomena coupling several numerical calculation codes. In the hope to obtain more precise results, a new type of numerical simulation appeared which simulate several physics at the same time. This type of application is called "codes coupling". In fact, several codes (physics) are coupled or interconnected so that they communicate to achieve the simulation.
This thesis is interested to the problematic of high performance coupling of parallel and distributed codes. Performance achievement is based on the conception of distributed applications where certain components are parallel and communicate eficiently. The basic idea of this thesis is to construct distributed application using a data-flow oriented parallel programming language (Athapascan). Our contribution is situated in two programming models ; "remote procedure call (RPC) model" and "stream-oriented model". The contributions brought by this research are the following ones :

- Utilisation of a data-flow oriented language on a RPC-based computational grid ;
In HOMA project, the extensions to RPC model carried on one hand on the control and communication semantic and on the other hand on the execution supports to better exploit the parallelism. In the homogenous architecture such as clusters, the theoretical results of these extensions are presented in the form of an execution cost model for an implementation on CORBA distributed system using KAAPI ; the execution kernel of Athapascan parallel language. The experiences (elementary and on a real application) validated this cost model.

- Extensions to a shared memory model for codes coupling ;
In order to extend the consistency model of Athapascan virtual shared memory, we proposed "time collection". This notation allows to describe the stream-like semantics of access to shared data. Spatial distribution of shared data, such as needed for "data
parallel" codes, is described by "space collection". To specify the semantic associated to these new notations, we gave a new definition for shared data. Then in the framework of this definition, we defined three types of shared data ; "sequential", "time collection" and "space collection".
Complete list of metadatas

Cited literature [72 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00010971
Contributor : Hamid-Reza Hamidi <>
Submitted on : Thursday, December 8, 2005 - 1:25:47 PM
Last modification on : Friday, November 6, 2020 - 4:38:54 AM
Long-term archiving on: : Monday, September 20, 2010 - 1:27:39 PM

Identifiers

  • HAL Id : tel-00010971, version 2

Collections

Citation

Hamid-Reza Hamidi. Couplage à hautes performances de codes parallèles et distribués. Réseaux et télécommunications [cs.NI]. Institut National Polytechnique de Grenoble - INPG, 2005. Français. ⟨tel-00010971v2⟩

Share

Metrics

Record views

548

Files downloads

394