Design and Implementation of a MassivelyParallel Version of DIRECT

摘要

This paper describes several massively parallel implementations for a global search algorithm DIRECT.Two parallel schemes take different approaches to address DIRECT's design challenges imposed by memory requirementsand data dependency. Three design aspects in topology, data structures, and task allocation are compared indetail. The goal is to analytically investigate the strengths and weaknesses of these parallel schemes, identify severalkey sources of inefficiency, and experimentally evaluate a number of improvements in the latest parallel DIRECTimplementation. The performance studies demonstrate improved data structure efficiency and load balancing on a2200 processor cluster.