Scheduling dynamically spawned processes in MPI-2

  • Authors:

  • Márcia C. Cera;Guilherme P. Pezzi;Maurício L. Pilla;Nicolas Maillard;Philippe O. A. Navaux

  • Affiliations:

  • Universidade Federal do Rio Grande do Sul, Porto Alegre, Brazil;Universidade Federal do Rio Grande do Sul, Porto Alegre, Brazil;Universidade Católica de Pelotas, Pelotas, Brazil;Universidade Federal do Rio Grande do Sul, Porto Alegre, Brazil;Universidade Federal do Rio Grande do Sul, Porto Alegre, Brazil

  • Venue:
  • JSSPP'06 Proceedings of the 12th international conference on Job scheduling strategies for parallel processing
  • Year:
  • 2006

Quantified Score

Hi-index 0.00

Visualization

Abstract

The Message Passing Interface is one of the most well known parallel programming libraries. Although the standard MPI-1.2 norm only deals with a fixed number of processes, determined at the beginning of the parallel execution, the recently implemented MPI-2 standard provides primitives to spawn processes during the execution, and to enable them to communicate together. However, the MPI norm does not include any way to schedule the processes. This paper presents a scheduler module, that has been implemented with MPI-2, that determines, on-line (i.e. during the execution), on which processor a newly spawned process should be run, and with which priority. The scheduling is computed under the hypotheses that the MPI-2 program follows a Divide and Conquer model, for which well-known scheduling algorithms can be used. A detailed presentation of the implementation of the scheduler, as well as an experimental validation, are provided. A clear improvement in the balance of the load is shown by the experiments.