Skip to Main content Skip to Navigation

# Modèles quantitatifs d'algorithmes parallèles

Abstract : This thesis presents ANDES, a technique for quantitative modelling of parallel algorithms and programs. An ANDES model is a valued DAG composed of computation nodes. The precedence is modelled by arcs. Data flow is modelled through input/output logics. Still, ANDES allows the representation of some non-determinist characteristics like branching. Hierarchical and regular models can also be constructed. Some examples are given. ANDES is introduced after an overview of other modelling techniques available in the literature (e.g. GMB). The ANDES-C library is used to represent ANDES models which are indeed C programs. One of the advantage of this textual representation is the possibility of easy description of thousands of computation nodes. ANDES can be employed inside different contexts of performance evaluation, mainly in order to describe a workload. This workload model can be given, for example, to a simulator or to an analytical model (e.g., a queueing system). In this work, ANDES is used to generate a synthetic load to be executed on a real parallel machine. This environment for generation and execution is called ANDES-Synth. Inside this environment, it is also possible to model a parallel machine which is emulated'' on the real multiprocessor. We use ANDES-Synth for the evaluation of static mapping strategies (four greedy heuristics and two iterative algorithms). A clustering technique (used in the Pyrros tool) is employed in order to apply the mapping strategies to the ANDES models. -- Taille du fichier postscript : 1521,932 KB
Keywords :
Document type :
Theses
Complete list of metadata

https://tel.archives-ouvertes.fr/tel-00005103
Contributor : Thèses Imag <>
Submitted on : Wednesday, February 25, 2004 - 2:11:46 PM
Last modification on : Friday, November 6, 2020 - 4:05:27 AM
Long-term archiving on: : Friday, September 14, 2012 - 11:40:56 AM

### Identifiers

• HAL Id : tel-00005103, version 1

### Citation

Joao-Paulo Kitajima. Modèles quantitatifs d'algorithmes parallèles. Réseaux et télécommunications [cs.NI]. Institut National Polytechnique de Grenoble - INPG, 1994. Français. ⟨tel-00005103⟩

Record views

Files downloads