Skip to Main content Skip to Navigation
Theses

Algorithmique parallèle du texte : du modèle systolique au modèle CGM

Abstract : We have all the intuition that a work can be completed in much less time if it is distributed between several people or on several machines. This concept names the parallelism which can be defined as the state of what develops in the same direction or at the same time. It is naturally that this concept of parallelism was applied to computers. So it was possible to answer to the needs for power necessary to the realization of greedy projects in computing times and in memory size. Parallelism combined with efficient algorithmic makes it possible to save time in order to answer important needs as well as possible. It breaks with the traditional approach which consists in gaining speed by carrying out each operation more quickly, approach limited by the laws of physics. Thus, the concept of parallelism largely contributed to the multiplication of data-processing models.We will be interested in systolic model and coarse grained parallel model baptized (Coarse Grained Multicomputers). CGM was proposed by F. Dehne and al. and it has properties which make it very interesting from a practical point of view. It is adapted perfectly to the modelling of existing architectures for which the number of processors can be several thousands and the size of the data can reach several billion bytes. An algorithm developed for this model consists of local computations using, if possible, optimal sequential algorithms and communication rounds of which the number must be independent of the size of the data to treat. Thus, CGM is very interesting from an economic point of view. Indeed, this model is independent of real architectures and makes it possible to re-use efficient sequential algorithms, which makes it very portable. In this thesis we are interested in parallel string matching algorithms problems. These problems can improve data compression or be used in bio-data processing. Thus, we propose CGM solutions for the problems of research of the longest increasing subsequence, longest common subsequence to two words, longest repeated suffix ending in each character of a word and repetitions. For that, we started from existing systolic solutions which we adapted to CGM. The goal of this work is in fact double. On the one hand, we propose for the first time CGM solutions to these four problems. In addition, we show how systolic solutions can be derived in CGM algorithms. Indeed, many problems were studied on systolic architectures, i.e. dedicated machines, no reusable for other problems. CGM makes it possible to work with inexpensive and reusable machines. Moreover, the experience gained during this work enables us to have a good idea of systolic solutions adaptable to CGM. This could make it possible to consolidate the existing bridge between fine grained models and coarse grained models. We finish this thesis by a discussion on load balancing of the solutions suggested and on the predictivity of the adaptation of other systolic solutions to CGM.
Complete list of metadatas

Cited literature [119 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00008672
Contributor : Thierry Garcia <>
Submitted on : Thursday, March 3, 2005 - 11:51:16 PM
Last modification on : Thursday, January 11, 2018 - 6:19:29 AM
Long-term archiving on: : Friday, April 2, 2010 - 9:39:04 PM

Identifiers

  • HAL Id : tel-00008672, version 1

Collections

Citation

Thierry Garcia. Algorithmique parallèle du texte : du modèle systolique au modèle CGM. Calcul parallèle, distribué et partagé [cs.DC]. Université de Picardie Jules Verne, 2003. Français. ⟨tel-00008672⟩

Share

Metrics

Record views

261

Files downloads

681