Efficiency of master-process ideology for parallel realization of iterative solving procedures for linear algebraic system

The effectiveness of master-process ideology application for realization of parallel solving procedures for system of linear algebraic equations is considered. Researches were conducted for iterative algorithms of conjugate gradient and Jacobi methods. The sparse format RR(C)U for coefficient matrix was used. Researches were conducted using system of linear algebraic equations computed with method of finite elements for plane boundary elastic problem. “Acceleration” calculated as ratio of successive algorithm execution time to parallel algorithm execution time was used as a criterion of quantity analysis. Analysis was conducted for algebraic systems from 100 to 50000 equations using six core AMD® Phenom II X6 1075T processor based computer. Program realization of iteration algorithms was created using C#. Standard MPI 2.0 was used to provide parallel processes communication. On base of obtained results it is possible to make a conclusion that usage of master-process parallel architecture for conjugate gradient iterative algorithm results in insignificant, from ten to fifteen percents, reduce of acceleration value as regards realization without master-process. This effect is much more lower for Jacobi method. Taking into account structural convenience of master-process parallel architecture the conclusion of possibility of this program architecture solution for considered methods was made.