Massively Parallel SPMD Algorithm for Cluster Computing— Combining Genetic Algorithm with Uphill

Genetic Algorithm (GA), which borrows the idea of Darwinian principle of natural selection, is a powerful global search and optimization method. This paper presents a SPMD(Single Program Multiple Data) algorithm which combines GA with local searching algorithm – uphill. The hybrid parallel method not only improves the convergence of GA but also accelerates the convergence speed of GA. Approximate solutions can be found quickly for complex optimization problems and more precise solutions can also be found by employing the same algorithm to fine-tune the approximate solutions. GA is an inherently [4] parallel algorithm. The SPMD algorithm exploits the parallelism of GA , at the same time, overcomes the premature and poor convergence properties of GA. The algorithm is applied on typical multiple local minima functions, TSP problem and an engineering computation problem QCBED on our selfdeveloped cluster system THNPSC-1. Experiments show that the algorithm is robust and it can find high quality solution with high speed.