doi:10.1016/S0167-739X(01)00075-9 
Copyright © 2002 Elsevier Science B.V. All rights reserved.
Exploiting cluster networks for distributed object groups and collective operations
a Fraunhofer Gesellschaft, Institute for Computer Architecture and Software Technology (FIRST), Kekuléstrasse 7, 12489 Berlin, Germany
b University of Tsukuba, 1-1-1 Tennodai, Tsukuba 305-8573, Japan
c RWCP, Tsukuba Mitsui Building 16F, 1-6-1 Takezono, Tsukuba 305-0032, Japan
Available online 1 January 2002.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
T
(Topologies and Collections) is a template-based object platform that strongly supports distributed data-parallel programming by means of distributed object groups and collective operations. This paper introduces T
s basic concepts and provides a performance analysis for collective operations on various cluster architectures with several different networks.
Author Keywords: Cluster computing; Distributed object groups; Collective operations
Fig. 2. Creation of object groups.
Fig. 3. Reduction over a binary tree.
Fig. 4. Calculating prefix sums.
Fig. 5. The BinTree topology class.
Fig. 6. The GroupPtr class.
Fig. 7. The Reduce functor.
Fig. 9. Functor generators.
Fig. 10. PCC II, Myrinet.
Fig. 11. PCC II, Fast Ethernet.
Fig. 12. PCC II, parameter passing bandwidth.
Fig. 14. HTPC, Fast Ethernet.
Fig. 15. HTPC, Gigabit Ethernet.
Fig. 16. HTPC, parameter passing bandwidth.
Fig. 17. Score Cluster III, Myrinet.
Fig. 18. Score Cluster III, Fast Ethernet.
Fig. 19. Score Cluster III, Myrinet parameter passing bandwidth.
Fig. 20. Score Cluster III, Fast Ethernet parameter passing bandwidth.
Abstract
Purchase PDF (312 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (0)
Purchase PDF (450 K)
