A dynamic task allocation scheme in Beneš-type nonblocking networks

  • Tse-yun Feng
  • Yanggon Kim
Load Balancing and Performance Optimization
Part of the Lecture Notes in Computer Science book series (LNCS, volume 797)

Abstract

Due to the uniqueness of a path between a pair of input/output terminals in cube-type networks, conflicts in the network are the main bottleneck in the performance of a highly-parallel multiprocessor system. In order to provide more flexibilities in realizing conflict-free connection patterns for a permutation, we consider rearrangeable nonblocking networks with (2 log2 N−1) stages. Based on an N × N Beneš network, a new dynamic task allocation scheme is shown to avoid both an intertask conflict and an intratask conflict. In addition, this scheme guarantees the independent partitioning of a MIN for an individual task. Therefore, the hot spot problem of a task caused by nonuniform memory traffic does not affect the other tasks and the effects of switch and/or link faults in the network is confined to a single task.

References

  1. 1.
    Tuomenoksa, D.L., Siegel, H.J.: Task Scheduling on the PASM Parallel Processing System. IEEE Trans. Software Eng. 11 (1985) 145–157Google Scholar
  2. 2.
    Premkumar, U.V., Browne, J.C.: Resource Allocation in Rectangular SW Banyans. Proc. Int. Symp. on Computer Architecture (1982) 326–333Google Scholar
  3. 3.
    Jeng, M., Siegel, H.J.: A Distributed Management Scheme for Partitionable Parallel Computers. IEEE Trans. Parallel and Distributed System 1 (1990) 120–126Google Scholar
  4. 4.
    Chang, Y., Bhuyan, L.N.: Extending Multistage Interconnection Networks for Multitasking. Proc. Int. Conf. Parallel Processing (1992) 155–158Google Scholar

Copyright information

© Springer-Verlag 1994

Authors and Affiliations

  • Tse-yun Feng
  • Yanggon Kim
    • 1
  1. 1.The Pennsylvania State UniversityUniversity ParkUSA
  2. 2.NSF/CISE/CDAArlington

Personalised recommendations