An improved approximation algorithm for pipelined operator tree scheduling

Abstract

Pipelined operator tree (POT) scheduling is an important problem in the area of parallel query optimization. A POT is a tree with nodes representing query operators that can run in parallel and edges representing communication between adjacent operators that is handled by sending long streams of data in a parallel-pipelined fashion. The problem is to assign operators to processors so as to minimize the maximum processor load. Chekuri et al. developed two algorithms for this NP-hard problem with performance ratios 3.56 and (1+ε)2.87 respectively, where ε > 0 can be made arbitrarily small. We present a (2+ε ) -approximation algorithm.