Empirical Study for Communication Cost of Parallel Conjugate Gradient on a Star-Based Network

Conjugate Gradient is an iterative linear solver that is used in many scientific and engineering applications to solve a system of linear equations. However, Conjugate Gradient generates a heavy load of computation and therefore it slows the performance of the applications using it. In this paper, we conduct an empirical cost study of a parallel CG on our star-based network. We evaluate the communication overhead involved by a parallel CG. In particular, we derive network parameters; the Maximum Transfer Unit (MTU), that can contribute to the optimization of communication cost and to the reduction of the waiting overhead of the parallel algorithm.