Chapter

Advances in Informatics

Volume 2563 of the series Lecture Notes in Computer Science pp 336-350

Date:

TOPPER: An Integrated Environment for Task Allocation and Execution of MPI Applications onto Parallel Architectures

  • Dimitris KonstantinouAffiliated withComputing Systems Lab, Computer Science Division Department of Electrical & Computer Engineering, National Technical University of AthensZografou Campus
  • , Nectarios KozirisAffiliated withComputing Systems Lab, Computer Science Division Department of Electrical & Computer Engineering, National Technical University of AthensZografou Campus

* Final gross prices may vary according to local VAT.

Get Access

Abstract

Although the use of parallel computing systems has significantly expanded in the last years, the existence of many processing elements is not fully exploited, due to the interprocessor communication overhead. In this paper we present an integrated software environment for optimizing the performance of parallel programs on multiprocessor architectures. TOPPER can efficiently allocate the tasks of a parallel application on the various nodes of a multiprocessing machine, using several algorithms for task clustering, cluster merging and physical mapping. The programmer outlines the application’s task computation and communication requirements along with the multiprocessor network available in two similar graphs. TOPPER aims to minimize the application’s overall execution time, proposing an efficient task allocation. In the case of MPI programs, TOPPER proves more powerful, since the application is automatically executed on the target machine with the provided task mapping.