Priority-based scheduling and evaluation of precedence graphs with communication times

Abstract

The design of distributed schemes requires partitioning a computation into smaller modules and scheduling it over a number of processors that communicate by exchanging packets of messages. The computation is represented by a directed acyclic graph. The objective function is to minimize the overall computation time. Due to the dynamic nature of the inter-tasks communication, an accurate global estimation of their effect is hard to find. As a result, local scheduling heuristic are used. In this work, new global priority based scheduling algorithms are presented, based on the evaluation of tasks finish times in the reverse graph. Extensive testing is carried out to evaluate the performance of these heuristics. Analysis of the results shows the superiority of these heuristics over reported ones as shorter finish times can be achieved with a stable topology independent performance. Finally, a new low-cost general iterative improvement technique is presented. Experimental testing of this technique is conducted with an analysis of the process behavior.