Skip to Main content Skip to Navigation

Modélisation et analyse des performances de la bibliothèque MPI en tenant compte de l'architecture matérielle

Meriem Zidouni 1 
1 VASY - System validation - Research and applications
Inria Grenoble - Rhône-Alpes, LIG - Laboratoire d'Informatique de Grenoble
Abstract : The range of high-end servers designed and manufactured by Bull includes cache-coherent distributed shared memory (CC-DSM) multiprocessor systems equipped with an implementation of the Message Passing Interface (MPI) library for parallel programming. The evaluation of the performances of this implementation will enable, on the one hand, to make the right choices of the hardware architecture and of the software layer at design time and, will provide, on the other hand, the analysis elements necessary for understanding the experimental measures performed when the real machine is validated. In this thesis, we propose and develop a methodology allowing to evaluate the performances of MPI library primitives (send/receive and barriers) by taking into account the hardware architecture. This approach is based on formal methods and consists of three main phases : 1) modeling in the LOTOS language the hardware aspects (interconnection topology, cache coherency protocol) and the software aspects (MPI primitives and benchmark algorithms) ; 2) formal verification of the functional correctness of the model obtained; 3) performance evaluation after extending the model with quantitative information (data transfer latencies) by using numerical methods and simulation
Document type :
Complete list of metadata

Cited literature [94 references]  Display  Hide  Download
Contributor : Meriem Zidouni Connect in order to contact the contributor
Submitted on : Wednesday, October 13, 2010 - 9:29:07 PM
Last modification on : Wednesday, July 6, 2022 - 4:15:21 AM
Long-term archiving on: : Friday, January 14, 2011 - 2:40:20 AM


  • HAL Id : tel-00526164, version 1


Meriem Zidouni. Modélisation et analyse des performances de la bibliothèque MPI en tenant compte de l'architecture matérielle. Modélisation et simulation. Université Joseph-Fourier - Grenoble I, 2010. Français. ⟨tel-00526164⟩



Record views


Files downloads