NICE: Scheduling Communication on Distributed Memory Machines

SUMMARY
This software package schedules collective unstructured communication required for parallelization of a number of irregular and loosely synchronous applications.


Grid used for generation of the communication matrix

PARTICIPATING INSTITUTIONS
NPAC, Syracuse University
School of Information Studies, Syracuse University

KEY CONTACTS
Sanjay Ranka | ranka@top.cis.syr.edu | (315) 443-4457

IMPACT
By reducing the node/link contention, our methods can reduce the total amount of time required for communication. This improves the speedup achieved on parallel machines for a variety of applications.

PROJECT DESCRIPTION
A communication package, NICE, is designed to help users in scheduling message-passing requests on distributed-memory machines. This package schedules a batch of messages into a set of partial permutations and provides communication primitives to carry out the communication. NICE primitives generate communication schedules to minimize node contention and/or link contention.

Results:

Experimental results (time in milliseconds) on a 32-node CM-5.

: NS- No Scheduling. : NICE primitives.

REFERENCES
  1. Sanjay Ranka, Jhy-Chun Wang, and Geoffrey C. Fox. Static and Runtime Algorithms for All-to-Many Personalized Communications on Permutation Networks. Proceedings of the 1992 International Conference on Parallel and Distributed Systems, pages pp. 211-218, HsinChu, Taiwan, December 1992.

  2. Sanjay Ranka, Jhy-Chun Wang, and Manoj Kumar. Personalized Communication Avoiding Node Contention on Distributed Memory Systems. Proceedings of the 1993 International Conference on Parallel Processing , volume I, pages pp. 241-244, St. Charles, IL, August 1993.

  3. Jhy-Chun Wang, Tseng-Hui Lin, and Sanjay Ranka. NICE: Non-uniform Irregular Communication Exchange on Distributed Memory Systems. Technical Report SCCS-546, NPAC at Syracuse University, June 1993.

Northeast Parallel Architectures Center, Syracuse University, npac@npac.syr.edu
This page maintained by Sanjay Ranka, ranka@top.cis.syr.edu