Authors:
Laxmikant Kale, Robert Brunner, James Phillips and Krishnan Varadarajan
Parallel Programming Laboratory, Department of Computer Science, University of Illinois at Urbana-Champaign

3rd Workshop on Runtime Systems for Parallel Programming

Clusters of PCs are an attractive platform for parallel applications because of their cost effectiveness. We have implemented an interoperable runtime system called Converse on a cluster of Linux PCs connected by an inexpensive switched Fast Ethernet. This paper presents our implementation and its performance evaluation. We consider the question of the performance impact of using inexpensive communication hardware on real applications, using a large production-quality molecular dynamics program, NAMD, that runs on this cluster.

[postscript] [PDF] [bibtex] [text reference]