FTL: a multithreaded environment for parallel computation

  • Authors:
  • Ivan Kalas;Eshrat Arjomandi;Guang R. Gao;Bill O'Farrell

  • Affiliations:
  • Centre for Advanced Studies, IBM Canada Ltd., 21/894, 844 Don Mills Rd., North York, Ontario M3C 1W3;Department of Computer Science, York University, North York, Ontario, M3J-1P3;-;Centre for Advanced Studies, IBM Canada Ltd., 21/894, 844 Don Mills Rd., North York, Ontario M3C 1W3

  • Venue:
  • CASCON '94 Proceedings of the 1994 conference of the Centre for Advanced Studies on Collaborative research
  • Year:
  • 1994

Quantified Score

Hi-index 0.00

Visualization

Abstract

The arrival of high-performance "killer micros" and the availability of high-performance networks (e.g., ATM) offer potential for building clusters of workstations with a significantly higher level of scalability than before. A promising approach to exploiting parallel computation on these systems is to use multithreading to overlap computation and communication while offering a simple programming model that smoothly integrates these two functions.This paper describes the design and implementation of a portable software platform for multithreaded computation in distributed memory systems. The goal is to provide a runtime environment that efficiently integrates computation and communication, and runs on off-the-shelf workstations without any hardware or operating system modifications. The target configurations are networked clusters of UNIX workstations, such as workstation farms and high-speed interconnect clusters. The FTL software platform is being implemented as a runtime library that can be used either directly by a programmer, or by a compiler. Portability and programmability are among the important objectives in our design.