Compile-time Priority Assignment and Re-routing for Communication Minimization in Parallel Systems

The performance gains of massively parallel systems can be significantly diminished by the inherent communication overhead. This overhead is caused by the required message passing resulting from the task allocation scheme. To minimize this overhead, a hybrid static-dynamic scheduling technique is presented. The static phase makes use of a priori information at compile-time to assign priorities to each message transmission. The priorities are determined using the recently developed Collision Graph model and are utilized at run-time to arbitrate the message transmissions. Determining an optimal priority scheme is an NP-Complete problem. Therefore the developed techniques employ heuristics and a flexible routing scheme to deal with a general case model of message traffic. Experiments performed show a significant improvement over baseline approaches.