Can Scatter Communication Take Advantage of Multidestination Message Passing?

  • 1Mohammad Banikazemi
  • Dhabaleswar K. Panda
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 1970)

Abstract

Collective communications compose a significant portion of the communications in many high performance applications. The multi-destination message passing mechanism has been proposed recently for efficient implementation ofcollectiv e operations by using a fewer number of communication phases. However, this mechanism has been only employed for improving the performance of non-personalized collective operations such as multicast. In this paper, we investigate whether multi- destination message passing can also help personalized communications such as one-to-all scatter. We propose a new scheme, called Sequential Multidestination Tree-based (SMT) scheme, which takes advantage ofthe multidestination message passing mechanism and provide a better per- formance. For a range of system sizes and parameters, it is shown that the SMT scheme outperforms other known schemes for a wide range of message lengths.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. 1.
    M. Banikazemi and D.K. Panda. Can Scatter Communication Take Advantage ofMultidestination Message Passing?. Technical Report OSU-CISRC-8/00-TR19, Dept. of Computer Science, The Ohio State University, 2000.Google Scholar
  2. 2.
    D. Dai and D. K. Panda. Reducing Cache Invalidation Overheads in Wormhole DSMs Using Multidestination Message Passing. In International Conference on Parallel Processing, pages I:138–145, Chicago, IL, Aug 1996.Google Scholar
  3. 3.
    L. Ni and P. K. McKinley. A Survey ofW ormhole Routing Techniques in Direct Networks. IEEE Computer, pages 62–76, Feb. 1993.Google Scholar
  4. 4.
    D. K. Panda. Issues in Designing Effcient and Practical Algorithms for Collective Communication in Wormhole-Routed Systems. In ICPP Workshop on Challenges for Parallel Processing, pages 8–15, 1995.Google Scholar
  5. 5.
    D. K. Panda, D. Basak, D. Dai, R. Kesavan, R. Sivaram, M. Banikazemi, and V. Moorthy. Simulation of Modern Parallel Systems: A CSIM-based approach. In Proceedings of the 1997 Winter Simulation Conference (WSC’97), pages 1013–1020, December 1997.Google Scholar
  6. 6.
    D. K. Panda, S. Singal, and R. Kesavan. Multidestination Message Passing in Wormhole k-ary n-cube Networks with Base Routing Conformed Paths. IEEE Transactions on Parallel and Distributed Systems, 10(1):76–96, Jan 1999.CrossRefGoogle Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2000

Authors and Affiliations

  • 1Mohammad Banikazemi
  • Dhabaleswar K. Panda
    • 1
  1. 1.Ohio State UniversityUSA

Personalised recommendations