Documents disponibles écrits par cet auteur

TASK. Partition based parallel design and implementation of the MSSM algorithm on a network of transputers / Cheref, Mohamed 

Public ISBD Titre : TASK. Partition based parallel design and implementation of the MSSM algorithm on a network of transputers Type de document : texte imprimé Auteurs : Cheref, Mohamed, Auteur ; Bouklachi, A., Directeur de thèse Editeur : Institut National d'Electricité et d'Electronique INELEC Année de publication : 1995 Importance : 115 f. Présentation : ill. Format : 30 cm. Note générale : Mémoire de Magister : Électronique : Boumerdès, Institut National d’Électricité et d’Électronique : 1995 Bibliogr. [4] f Langues : Anglais (eng) Mots-clés : Network  of  transputers  ;  Parallel  systems  taxonomies  ;  Parallel  processing  systems  programming  ;  Transputers  ;  MSSM  algorithm  ;  Multiple  scale  signal  matching  algorithm  ;  Sequential  design  and  implementation  ;  Parallel  design  and  implementation Index. décimale : M004995 Résumé : The present work is concerned with the parallel design and implementation of the Multiple Scale Signal Matching (MSSM)algorithm on a transputer network. The MSSM algorithm is based on a multichannel vision model, to establish the correspondence between two images with the allowance that one of them can be deformed elastically. The MSSM algorithm uses a process consisting of two stages: the filtering and the matching stages. This process is iterated following a coarse-to-fine regime of the vision channels at which the matching process is performed. Therefore, the algorithm exhibits a certain degree of computational complexity over huge amounts of data. This suggests the use of parallel processing to reduce the execution time of the algorithm. For this reason we have considered the parallelization of the algorithm over a PC transputer network with the OCCAM 2 language under the Transputer Development System TDS3. A task partition approach is used to parallelize the algorithm. First, the algorithm is partitioned into a set of elementary tasks. Then, an intertask data flow is established. Afterwards, the network topology is fixed and subsequently the tasks are placed onto the transputer network processors. Finally, the tasks are scheduled onto processors in order to minimize the processors idle time caused by the intertask data flow, and obtaining an optimal starting time for each task ignition. The experiments were carried out to measure the performance of the parallel implementation with respect to the sequential one. The effect of the increase in the computational time when additional vision channels are used, with respect to the increase in the number of processors has been also investigated. The performance tests indicate a substantial improvement in speed compared with a single processor execution. The performance analysis has permitted us to identify the optimal number of processors suitable for such an application. TASK. Partition based parallel design and implementation of the MSSM algorithm on a network of transputers [texte imprimé] / Cheref, Mohamed, Auteur ; Bouklachi, A., Directeur de thèse . - [S.l.] : Institut National d'Electricité et d'Electronique INELEC, 1995 . - 115 f. : ill. ; 30 cm. Mémoire de Magister : Électronique : Boumerdès, Institut National d’Électricité et d’Électronique : 1995 Bibliogr. [4] f Langues : Anglais (eng) Mots-clés : Network  of  transputers  ;  Parallel  systems  taxonomies  ;  Parallel  processing  systems  programming  ;  Transputers  ;  MSSM  algorithm  ;  Multiple  scale  signal  matching  algorithm  ;  Sequential  design  and  implementation  ;  Parallel  design  and  implementation Index. décimale : M004995 Résumé : The present work is concerned with the parallel design and implementation of the Multiple Scale Signal Matching (MSSM)algorithm on a transputer network. The MSSM algorithm is based on a multichannel vision model, to establish the correspondence between two images with the allowance that one of them can be deformed elastically. The MSSM algorithm uses a process consisting of two stages: the filtering and the matching stages. This process is iterated following a coarse-to-fine regime of the vision channels at which the matching process is performed. Therefore, the algorithm exhibits a certain degree of computational complexity over huge amounts of data. This suggests the use of parallel processing to reduce the execution time of the algorithm. For this reason we have considered the parallelization of the algorithm over a PC transputer network with the OCCAM 2 language under the Transputer Development System TDS3. A task partition approach is used to parallelize the algorithm. First, the algorithm is partitioned into a set of elementary tasks. Then, an intertask data flow is established. Afterwards, the network topology is fixed and subsequently the tasks are placed onto the transputer network processors. Finally, the tasks are scheduled onto processors in order to minimize the processors idle time caused by the intertask data flow, and obtaining an optimal starting time for each task ignition. The experiments were carried out to measure the performance of the parallel implementation with respect to the sequential one. The effect of the increase in the computational time when additional vision channels are used, with respect to the increase in the number of processors has been also investigated. The performance tests indicate a substantial improvement in speed compared with a single processor execution. The performance analysis has permitted us to identify the optimal number of processors suitable for such an application.

Exemplaires

Code-barres	Cote	Support	Localisation	Section	Disponibilité	Spécialité	Etat_Exemplaire
M004995	M004995	Papier	Bibliothèque centrale	Mémoire de Magister	Disponible

Documents numériques

CHEREF.Mohamed.pdf URL