Communication/computation paradigms for hypercubes

Authors:
R. E. Horan;D. A. Poplawski
Affiliations:
Department of Computer Science, Mitchigan Technological University, Houghton, Michigan;Department of Computer Science, Mitchigan Technological University, Houghton, Michigan
Venue:
C3P Proceedings of the third conference on Hypercube concurrent computers and applications: Architecture, software, computer systems, and general issues - Volume 1
Year:
1988

Citing 2
Cited 1

Computer networks

Computer networks
Computer Architecture and Parallel Processing

Computer Architecture and Parallel Processing

A flexible system call interface for interprocessor communication in a distributed memory multicomputer

ACM SIGOPS Operating Systems Review

Quantified Score

Hi-index	0.00

Visualization

Abstract

Most hypercube node operating systems allow multiple (multiprogrammed) processes per node and buffered, asynchronous communication with routing between arbitrary nodes. The FPS T-Series implements the CSP (communicating sequential process) model of computation on each node and unbuffered, synchronous communication between adjacent nodes. We have implemented on the T-Series several other computation/communication paradigms (for example, single process with routed, unbuffered, synchronous communication) and measured many aspects of communication performance and the computation overheads associated with each. In this paper we compare the performance and overheads of these and various other models of computation/communication and discuss the various models' flexibility and usefulness in typical applications.