Research Of Parallel Program Performance Visualization Under Heterogeneous Platforms

In recent years, with the growing demand for computational, parallel clusters and even the growing multi-core cluster, makes high-performance parallel computing more widely used in large areas of data processing. However, there is unity and the lack of useful graphical tools deficiencies of PC cluster nodes and cluster computing operating system, parallel computing allows users to relatively complex operation, not intuitive grasp of the program is running, identify performance bottlenecks, not conducive to the performance indicators in accordance with procedures designed to deficiencies found in. Therefore, in order to facilitate the programmers and users to access, generated performance information analysis program, visualization technology as a observation, effective way of program performance gradually into the field of parallel computing, and have attracted much attention by more and more people.In order to better grasp the cluster system as performance parallel applications, this paper studies the context of related work at home and abroad constructed based on MPI and MPI-TBB model of two parallel programming parallel Linux computing cluster, and control displays the user with the windows platform together to form a heterogeneous environment, and on this basis gives a visual model of parallel computing performance.In this paper, parallel program execution time, CPU utilization and the implementation process of the various states and the performance event to detection program performance situation, proposed to use code instrumentation, MPE detection and logging methods accomplish forward visualization of parallel computing performance. Design of C # language integrated terminal emulation software and log analysis tools; performance through a system enables users to quickly view, intuitive way to learn about parallel program execution, communication situation. To locate performance bottlenecks, optimization program provides visual reference. In order to verify the feasibility and effectiveness of this article, MPI programming model through the ranks of matrix multiplication parallel program with MPI-TBB hybrid programming model to test the queens'placement problem, respectively show different aspects of performance procedures. Experiments show that, visualization methods in this paper has the characteristics of practicality and effectiveness, cluster system to improve the availability, performance and efficiency improvement program has good practical value.