Irregular computations on fine-grain multithreaded architecture

Parimala Thulasiraman

ISBN 10: 3838322142 / ISBN 13: 9783838322148

Published by LAP LAMBERT Academic Publishing Okt 2009, 2009

About this Item

Neuware -Engineering and other elds require high performance parallel computing. When the problem at hand presents a structure regular enough to be dealt with automatically, it is sufficient to leave the parallelization and optimization to compiling tools. However, most of the important applications possess structures that lack regularity, and consequently, performance enhancement by exploitation of high performance computing requires new design techniques and tools. In these cases, issues such as data layout, data dependencies,locality, workload, load balance and minimization of communication which impact performance need to be carefully assessed for the client parallel implementation. Memory and synchronization latencies are serious bottlenecks in parallel computing. In the SPMD (Single Instruction Multiple Data Model), von Neumann style of programming is used and hence irregular computations are hard to program. On the other hand, in negrained multithreading models, the program is divided into many small threads and the threads interleaved to achieve maximum parallelism.Books on Demand GmbH, Überseering 33, 22297 Hamburg 228 pp. Englisch.

Seller Inventory # 9783838322148

Bibliographic Details

• Title: Irregular computations on fine-grain multithreaded architecture
• Publisher: LAP LAMBERT Academic Publishing Okt 2009
• Publication Date: 2009
• Language: English
• ISBN 10: 3838322142
• ISBN 13: 9783838322148
• Binding: Taschenbuch
• Condition: Neu
• Dimensions: 220W x 150H x 14D millimeters 220 millimeters width by 150 millimeters height by 14 millimeters depth
• Weight: 358 grams

About this title

Synopsis

Engineering and other elds require high performance parallel computing. When the problem at hand presents a structure regular enough to be dealt with automatically, it is sufficient to leave the parallelization and optimization to compiling tools. However, most of the important applications possess structures that lack regularity, and consequently, performance enhancement by exploitation of high performance computing requires new design techniques and tools. In these cases, issues such as data layout, data dependencies,locality, workload, load balance and minimization of communication which impact performance need to be carefully assessed for the client parallel implementation. Memory and synchronization latencies are serious bottlenecks in parallel computing. In the SPMD (Single Instruction Multiple Data Model), von Neumann style of programming is used and hence irregular computations are hard to program. On the other hand, in negrained multithreading models, the program is divided into many small threads and the threads interleaved to achieve maximum parallelism.

About the Author

Parimala Thulasiraman received B.Eng.(Honors)and M.A.Sc.degrees in Computer Engineering from Concordia University, Montreal, Canada and obtained her Ph.D. from University of Delaware, Newark,DE, USA after finishing most of her formalities in McGill University, Montreal, Canada. She is now an Associate Professor at Manitoba University.