A Framework for Collective Personalized Communication
IEEE International Parallel and Distributed Processing Symposium (IPDPS) 2003
Publication Type: Paper
Repository URL: CommLib
This paper explores collective personalized communication. For example, in all-to-all personalize d communication (AAPC), each processor sends a distinct message to every other processor. However, for many applica tions, the collec- tive communication pattern is many-to-many, where each processor sends a distinct message to a subs et of processors. In this paper we first present strategies that reduce per-message cost to optimize AAPC. We then prese nt performance results of these strategies in both all-to-all and many-to-many scenarios. These strategies are implemented in a flexible, asynchronous library with a non-blocking interface, and a message-driven runtime system. This allows the collect ive communication to run concurrently with the application, if desired. As a result the computational overhead of the commun ication is substantially reduced, at least on machines such as PSC Lemieux, which sport a co-processor capable of remote DMA . We demonstrate the advantages of our framework with performance results on several benchmarks and applications.
L. V. Kale and Sameer Kumar and Krishnan Vardarajan, "A Framework for Collective Personalized Communication, Communicated to IPDPS 2003", Parallel Programming Laboratory, Department of Computer Science, University of Illinois at Urbana-Champaign, 2002.
Research Areas