Comparative Analyses of Parallel Simulation Protocols

Report
Authors:Reynolds, Jr, Institute for Parallel ComputationUniversity of Virginia Weight, C, Institute for Parallel ComputationUniversity of Virginia Fidler, II, Institute for Parallel ComputationUniversity of Virginia
Abstract:

Currently there is not a significant body of comparative, experimental performance results for parallel simulation protocols. Nor is there a body of significant analytic studies. The SPECTRUM Testbed [ReDi89] has been designed to support the empirical study of parallel simulation protocols and applications, with the expectation that experience with the testbed will provide insights into the efficacy of various protocols and their interplay with classes of applications. We discuss our experience with the SPECTRUM Testbed, focusing primarily on an observed, unexpected degree of dependency between protocols and applications, and on an unexpected, large set of application design options. The latter gives rise to the definition of a set of application design variables which describes a large design space. We discuss its impact on the testbed design and we discuss a limited set of performance results that we have for selected sets of protocols and applications.
Note: Abstract extracted from PDF text

Rights:
All rights reserved (no additional license for public reuse)
Language:
English
Source Citation:

Reynolds, Jr, C Weight, and II Fidler. "Comparative Analyses of Parallel Simulation Protocols." University of Virginia Institute for Parallel Computation Tech Report (1989).

Publisher:
University of Virginia, Institute for Parallel Computation
Published Date:
1989