How to get good performance from the CM-5 data network

摘要

Programmers of the Connection Machine CM-5 data network can improve the performance of their data movement code more than a factor of three by selectively using global barriers, by limiting the rate at which messages are injected in to the network, and by managing the order in which they are injected. Barriers eliminate target-processor congestion, and allow the programmer to schedule communications globally. Injection-reordering improves the statistical independence of the various packets in the network at any given time. Barriers and tuned injection rates provide forms of flow control. Barriers also provide a composition of performance property: if you understand the performance of parallel computations A and B, then you understand the performance of "A; barrier; B". Architectural support for global barriers, injection reordering, and flow control may be worthwhile for achieving good communications performance. Although our evidence comes from the CM-5, we expect these techniques to apply to most parallel machines.