| With the continuous development of science and technology, the performance of a single core processor by cooling, power consumption and leakage current and a series of factors, has been difficult to have bigger breakthrough. Multi-core processor arises at the historic moment and has become one of the hottest research topic today. OpenMP as the DE facto standard Shared storage model, because compared to write parallel programs is simple, and easy to implement on multi-core processors, the parallel program writers love, and be recognized by more and more people.In this paper, starting from the parallel technology of multi-core processors, first of all,the parallel computation, the key technology of this technology are expounded and the research, mainly analyzes on the advantages of the parallel programming and the difficulties of development of popularization. And then studied the two basic programming model of parallel programming:the MPI and the OpenMP. MPI based on message passing is easy to extension but is difficult to write programs, however, OpenMP is easy programming, but not easy to expand, and need to be synchronized mechanism. And then multicore multi-threaded scheduling algorithm is of further research based on OpenMP. In OpenMP, in view of the for loop scheduling, there are four ways:static, dynamic, guided the runtime.They are to control each thread loop iteration number when performing a for loop to achieve the balance between the scheduling overhead and load balancing. At the end of the paper, after analysis research, proposed the key of improvements on scheduling algorithm is the iteration number of every time each thread scheduling loop, and according to the directions, this paper proposes a new scheduling strategy of combination of dynamic and guided scheduling, and is tested by experiment, the results show that the new overall performance is superior to the guided and static scheduling strategy, and is equal to the dynamic. |
PDF Full Text Request