Exploiting heterogeneous parallelism in the presence of communication delays

The interprocessor communication overhead in distributed memory parallel architectures can have a significant impact on overall machine performance, so scheduling techniques that can be used to reduce this overhead are important. In this paper, we design a new priority-based scheduling algorithm in an attempt to maximize heterogeneous parallelism while minimizing communication delays, based on the concept of the increase over the execution time of a given program. To estimate the goodness of the algorithm by relating its performance to an optimal solution, we also develop a branch-and-bound algorithm using a new lower bound cost function. Finally, we present experimental results to demonstrate the effectiveness and efficiency of the optimization technique.