Colloquium Details
Global Data Distribution in Software Distributed Shared Memory Systems
| Author: | David Lowenthal Department of Computer Science, University of Georgia |
|---|---|
| Date: | March 15, 2001 |
| Time: | 15:30 |
| Location: | 220 Deschutes |
Abstract
Distributing data is one of the key problems in implementing efficient distributed-memory parallel programs. The problem becomes more difficult in programs where data redistribution between computational phases is considered. The global data distribution problem is to find the optimal distribution in multi-phase parallel programs. We are investigating this problem in the context of a software distributed shared memory (SDSM) system. In this talk, we first describe SUIF-Adapt, which is an integrated compiler/run-time system for finding efficient global data distributions, and then discuss performance of SUIF-Adapt programs. We next describe how SUIF-Adapt obtains accurate redistribution time estimates. Obtaining these estimates is difficult because it depends on access patterns, page locations, and the SDSM consistency protocol. However, having accurate redistribution times is critical to choosing an efficient distribution. We discuss our integrated compiler/run-time method for finding accurate time estimates and describe optimizations that improve the performance of our algorithm.