Some Parallel Programming Techniques Based On MPI

Because of the development of technology, the new generation of computer should be better than the former ones in the power and speed of computing. But the people's demand of high performance of computing is increasing too. In addition to enhancing the computing power of the processor, parallel processing is also an efficient way to enhance the power of computing. In the past, the parallel processing can only run on the expensive and special computers. As the cost of personal computers and networks decreased, and now, it is popular to process the parallel processing on the distributed network computing systems. In the distributed network computing systems, message passing is used for the communication between processes. MPI (Message Passing Interface) and PVM (Parallel Virtual Machine) are common development environments of parallel processing based on message passing. This article briefs the background of MPI and its functions, including communicator, process group, blocking and non-blocking communications etc. MPI has been compared with PVM in aspects of the purposes of their developments, portability and interoperability, process group and communications etc. Based on the MPI, we studied the parallel programming techniques of SPMD of cluster. The MPI frameworks of the functions of data partition, load balance, avoiding dead-lock, non-blocking communication are given. Finally, for reference, the solutions of the N-queen and Travelling Salesman are given to the programmers and students, who are studying and developing to parallel programming of MPI. The parallel computing environment used by examples, is based on the implementation of LAMIMPI of Linux platform.