Design and implementation of a performance analysis and visualization toolkit for cluster environments

  • Authors:

  • Tien-Hsiung Weng;Hsiao-Hsi Wang;Tsung-Ying Wu;Ching-Hsien Hsu;Kuan-Ching Li

  • Affiliations:

  • Parallel and Distributed Processing Center, Dept. of Computer Science and Information Engineering, Providence University, Shalu, Taichung, Taiwan;Parallel and Distributed Processing Center, Dept. of Computer Science and Information Management, Providence University, Shalu, Taichung, Taiwan;Grid Operation Center, National Center for High Performance Computing, Taichung City, Taichung, Taiwan;Dept. of Computer Science and Information Engineering, Chung Hua University, Hsinchu, Taiwan;Parallel and Distributed Processing Center, Dept. of Computer Science and Information Engineering, Providence University, Shalu, Taichung, Taiwan

  • Venue:
  • ICHIT'06 Proceedings of the 1st international conference on Advances in hybrid information technology
  • Year:
  • 2006

Quantified Score

Hi-index 0.00

Visualization

Abstract

The low cost and wide availability of PC-based clusters have made them excellent alternatives to supercomputing. However, while Network of Workstations are readily available, there is an increasing need for performance tools that support these computing platforms in order to achieve even higher performance. Strategies that may be considered toward such performance achievement we may list are: performance data analysis, algorithm design, parallel program restructuring, among others. Introduced in this paper is a toolkit that generates performance data and graphical charts of pure MPI, pure OpenMP, as well as hybrid MPI/OpenMP parallel applications, reflecting to its sequence of execution over time and cache behavior, with the use of DP*Graph representation, a parallel version of timing graph. That is, parallel applications have their execution sequence in a cluster system platform shown through graphical charts composed by sequential codes, parallel threads, dependencies and communication structures, symbols defined in DP*Graph. It is discussed the implementation of this toolkit, as also some of its features, together with experimental use of the toolkit on parallel applications such as matrix multiplication (parallel implementation using MPI) and SPICE3 (parallel implementation using OpenMP).