Divide-and-conquer techniques for global throughput optimization

Abstract

The paper proposes a divide-and-conquer approach for global throughput optimization designed to coordinate existing techniques and enable their more effective use. The "divide" approach consists of logical partitioning of the computation into subparts. The techniques for partitioning the computation, and the corresponding scheme for classifying the subparts is presented. The subparts are optimized or "conquered" through coordinated application of existing optimization techniques. Optimization techniques that are effective for each class have been characterized in terms of their expected effect on throughput. The approach is not limited to a specific class of computations and gives higher, or at least equal, improvement than previously reported techniques on all examples.