Optimizing message delivery in asynchronous distributed applications

Abstract

Since the message delivery time of asynchronous applications is unpredictable, each asynchronous process must probe (or poll) the network for new messages regularly. If polling is carried out too aggressively, a majority of the probes will be unsuccessful. However, the cost for probing the network is similar to the cost of communication operations. Thus, optimizing the message reception behavior has a significant impact on the performance of the application. This paper studies this problem, and develops a cost model to optimize the polling frequency. We also develop strategies to optimize the message delivery behavior, and study their impact on the performance of a parallel discrete event simulator. © Springer-Verlag Berlin Heidelberg 1999.