Abstract
New generations of scientific codes trend to mix different types of parallelism. Algorithms are defined as a set of modules, with data parallelism inside modules and task parallelism between them. With high speed networks, tasks running on a heterogeneous computing environment can exchange data in a reasonable delay. Therefore data-parallel tasks distributed on different parallel computers can interact efficiently by reading or writing Data Parallel Objects. These objects are distributed on the physical nodes according to the mapping directives. Migrations of data parallel objects from one parallel computer to another lead us to define efficient algorithms for runtime array redistribution. In this work, we have specially cared about the ability to handle distinct source and target processor sets while performing redistribution and the ability to overlap communications and computations. Performance results on a farm of ALPHA processors are discussed.
Chapter PDF
Similar content being viewed by others
Keywords
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
References
B. Avalani, A. Choudhary, I. Foster, R. Krihnaiyer, and M. Xu. A data transfer library for communicating data-parallel tasks. Technical report, Syracuse University, Argone Natinal Laboratory, NY 13244, 1994.
D. A. Carlson. Ultrahigh-performance FFTs for the CRAY-2 and CRAY-YMP supercomputers. The journal of supercomputing, 6(2):107–116, june 1992.
G. Edjlali, N. Emad, and S. Petiton. Hybrid methods on network of heterogeneous parallel computers. Proceedings of the 14th Imacs world congress Atlanta, USA, july 1994.
D. Sueur. Shell hétérogène à parallélisme de données. In Renpar'7, Actes des 7 Rencontres Francophones du parallélisme, pages 58–61, PIP-FPMs Mons, Belgique, June 1995.
Author information
Authors and Affiliations
Editor information
Copyright information
© 1996 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Sueur, D., Luc-Dekeyser, J. (1996). Dynamic redistribution on heterogeneous parallel computers. In: Bougé, L., Fraigniaud, P., Mignotte, A., Robert, Y. (eds) Euro-Par'96 Parallel Processing. Euro-Par 1996. Lecture Notes in Computer Science, vol 1123. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-61626-8_22
Download citation
DOI: https://doi.org/10.1007/3-540-61626-8_22
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-61626-9
Online ISBN: 978-3-540-70633-5
eBook Packages: Springer Book Archive