RANDOMIZED OPTIMAL LIST RANKING ON COARSE-GRAINED PARALLEL COMPUTERS WITH O(log p) COMMUNICATION PHASES

Abstract

The cost of interprocessor communication has a substantial impact on execution time when implementing parallel algorithms on physical parallel computers. For coarse-grained parallel computers it is important to minimize the number of communication phases, in order to balance cost of communication with local computation time. We present a log p-phase optimal randomized parallel list ranking algorithm and its application to expression evaluation. These techniques address the general issue of model-independent parallel algorithm design.