Parallel selection by regular sampling

Abstract

Bulk-synchronous parallelism (BSP) is a simple and efficient paradigm for parallel algorithm design and analysis. In this paper, we present a new simple deterministic BSP algorithm for the classical problem of selecting the k-th smallest element from an array of size n, for a given k, on a parallel computer with p processors. Our algorithm is based on the technique of regular sampling. It runs in optimal O(n/p) local computation and communication, and near-optimal O(loglogp) synchronisation. The algorithm is of theoretical interest, as it gives an improvement in the asymptotic synchronisation cost over its predecessors. It is also simple enough to be implementable. © 2010 Springer-Verlag.