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 Locality Management ONR/ARPA contract N00014-92-J-1801

Locality Management in Large-Scale Multiprocessors

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.

Principal Investigators

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

Graduate Students

Recent Graduates

Recent Accomplishments

Publications

The Mint Multiprocessor Simulator

MINT is a software package designed to ease the process of constructing event-driven memory hierarchy simulators for multiprocessors. It provides a set of simulated processors that run standard Unix executable files compiled for a MIPS R3000 based multiprocessor. These generate multiple streams of memory reference events that drive a user-provided memory system simulator. MINT uses a novel hybrid technique that exploits the best aspects of native execution and software interpretation to minimize the overhead of processor simulation. Combined with related techniques to improve performance, this approach makes simulation on uniprocessor hosts extremely efficient.

Mint is available without charge to academic and research sites.


ARPA project page

Last Change: 1 March 1995 / scott@cs.rochester.edu