Date: Wednesday, 20-Nov-96 20:04:52 GMT Server: NCSA/1.3 MIME-version: 1.0 Content-type: text/html Last-modified: Friday, 23-Aug-96 20:29:29 GMT Content-length: 4004
Computer Science Department
University of Rochester
Rochester, NY 14627-0226
6-1-92 through 5-31-95
This project is developing techniques that improve locality of reference in parallel programs, so as to admit efficient execution on large-scale multiprocessors. We use a combination of simulation, analytic studies, and experimentation to compare alternative techniques for locality management under various architectural assumptions, using a wide variety of benchmark applications. We seek to quantify the performance impact of individual techniques at the architecture, operating system, runtime library, and user level, and the interaction between techniques across layers in the system. This work will lead to a better understanding of the role of locality and data sharing in parallel program performance, and a comprehensive understanding of the effectiveness of techniques for locality management and their interactions.
Thomas J. LeBlanc
Professor and Chair
leblanc@cs.rochester.edu
716-275-5478
Michael L. Scott
Associate Professor
scott@cs.rochester.edu
716-275-7745
Mint is available without charge to academic and research sites.
Last Change: 1 March 1995 / scott@cs.rochester.edu