Parallel Computing Research Based On The High Performance Computer

Parallel computing is a hot issue in the field of computer science, many countries are scrambling to invest heavily in the development of high-performance computing systems. In this paper, mainly study on parallel computing power of a single node in cluster, the mainly finished research works are as follows:The characteristics of 64-bit microprocessor and the use of SIMD instructions are firstly studied, then introduce the steps and methods to build 64-bit development environment. Briefly introduce the use of SIMD instructions, show speedup via a Quantitative Analysis on a image processing program using SIMD instructions.Through analysis and research on a variety of parallel hardware platforms, use method of multi-thread programming, analysis the two methods of seperating data by rows or columns. And use the method on-row block, propose a variety of synchronization solution and test on parallel multicore platform, hyper-threading parallel platform, multiprocessor parallel platform to compare their performance, and put forward a way to reduce synchronization and improve the parallelism of the method. Via the analysis of experimental data summarized, put forward to the general steps about multithreaded parallel programming method. The experiment results show that the multi-processor architecture suitable for a continuous block by row, while multi-core and hyper-thread parallel architecture suitable for a cross block by row.