Towards a Theory of Optimal Communication Pipelines

In this paper, we study how to minimize the latency of a message through a network that consists of a number of store-and-forward stages. This research is especially relevant for today's low overhead communication subsystems that employ dedicated processing elements for protocol processing. We develop an abstract pipeline model that reveals a crucial performance tradeo . We subsequently exploit this tradeo and present a series of fragmentation algorithms designed to minimize message latency. We provide an experimental methodology that enables the construction of customized pipeline algorithms that can adapt to the speci c pipeline characteristics and application workloads. By applying this methodology to the Myrinet-GAM system, we have improved its latency by up to 51%. We also study the e ectiveness of this technique for other realistic cases.