Parallel Optimization and Performance Tuning on a Kunpeng Cluster of Genetic Algorithm for Synthesis of Circulant Networks

Parallel versions of a genetic algorithm based on the hybrid MPI—OpenMP model are implemented to optimize circulant networks, which are of practical interest in the design of supercomputer systems and systems on a chip. An analysis of the efficiency of parallel programs with different numbers of MPI processes and OpenMP threads on a cluster of Kunpeng processors has been carried out. The speed-up of several hybrid parallel computing schemes was experimentally evaluated and analyzed. Two bottlenecks in terms of efficiency in parallel execution of the algorithm are identified and methods for their solution are proposed. By means of the parallel genetic algorithm the descriptions of circulant networks with better average distance and bisection width for the known large circulant networks were obtained.