Cluster-based parallel computing is becoming the mainstream of high performance computing techniques, has been used more widely in science and industry. More and More engineers need to program on Cluster-based parallel computer by themselves. They hope to program correctly and effectively, find out the logic error quickly, and locate the bottlenecks in time, improve the performance of parallel program.RMAPCP (Real-time Monitor and Analysis Tools for Parallel Computing Performance) can help them to achieve the goals. RMAPCP real-time collects the performance information of the process of parallel program computing and the information of node's performance, shows them to programmer. By the aids of observing and analyzing the feedback information, the programmer who works with MPI can judge whether the process of parallel program is correct, and rectify the program, improve the performance of parallel program.The paper introduces the RMAPCP system's runtime environments, design background, the whole design thinks, and implement methods. It probes into the key technology of performance monitor functions, daemon process on computing node, and administrator front-end. The discussion mainly includes the collection and convey of information of performance and load, implement of the real-time monitor, the commutation of front-end and daemon process, the display and analysis of performance information. An example, which uses RMAPCP in parallel computing, is presented at last.
|