Randomized parallel communication (Preliminary Version)

  • Authors:

  • R. Aleliunas

  • Affiliations:

  • -

  • Venue:
  • PODC '82 Proceedings of the first ACM SIGACT-SIGOPS symposium on Principles of distributed computing
  • Year:
  • 1982

Quantified Score

Hi-index 0.01

Visualization

Abstract

Using a simple finite degree interconnection network among n processors and a straightforward randomized algorithm for packet delivery, it is possible to deliver a set of n packets travelling to unique targets from unique sources in 0(log n) expected time. The expected delivery time is in other words the depth of the interconnection graph. The b -way shufile networks are examples of such. This represents a crude analysis of the transient response to a sudden but very uniform request load on the network. Variations in the uniformity of the load are also considered. Consider si packets with randomly chosen targets beginning at a source labelled i. The expected overall delay is then [equation] where the labelling is chosen so that s1≥s2≥. These ideas can be used to guage the asymptotic efficiency of various synchronous parallel algorithms which use such a randomized communications system. The only important assumption is that variations in the physical transmission time along any connection link are negligible in comparison to the amount of work done at a processor.