Managing Data Synchronization Automatically for Distributed-Memory Architectures
Ko‐Yang Wang · Purdue e-Pubs (Purdue University System) · 1990
In this paper, we discuss problems in parallelizing sequential programs for distributed-memory parallel computers. In particular, we introduce a new program transfonnation technique called the message consolidation. The message consolidation reduces the communication time for parallel computers that are based on message passing paradigm by decreasing the number of messages on the communication network. Effects of the message consolidation on program performance and conditions for which it is applicable are analyzed. An algorithm for consolidating messages and some heuristics of recognizing the opportunities for applying the transformation are also presented.