Optimizing the algorithms for parallel computing systems requires effective dynamic load balancing. OpenMP and MPI technologies allow for software development for the most of modern parallel computing architectures but MPI does not provide standard facilities for dynamic load balancing. In this paper, we propose MPI-MAP parallel programming pattern, which solves that problem by implementing MAP stage of well-known MAP-REDUCE pattern with standard facilities of MPI. The efficiency of dynamic load balancing with OpenMP and MPI-MAP is shown on symmetric multiprocessing systems with shared memory, massive parallel processing clusters with distributed memory and computing systems with Intel Xeon Phi co-processors. As an example, the problem of numerical modeling the influence of typical nonlinearities on output vector functions of nonlinear stabilization system of mobile control object is solved.

  Повний текст PDF - 1.021 Mb    Зміст випуску     Цитування публікації

  Якщо, ви не знайшли інформацію про автора(ів) публікації, маєте бажання виправити або відобразити більш докладну інформацію про науковців України запрошуємо заповнити "Анкету науковця"