Scheduling parallel tasks on heterogeneous clusters

• Type: Article in International Conference Proceedings Book
• Date: 2004

Authors

Jorge Manuel Gomes Barbosa (Author) FEUP; Celeste Isabel Navarro Morais (Author) Other; Miguel Pimenta Monteiro (Author) FEUP

Conference proceedings International

Title: International Conference on Parallel and Distributed Computing and Systems

Pages: 1-6

IASTED

Boston, Agosto

Scientific classification

CORDIS: Technological sciences > Engineering

Other information

Abstract (EN): Parallel tasks, also called malleable tasks, are tasks that can be executed on any number of processors with its execution time being a function of the number of processors alloted to it. The scheduling of independent parallel tasks on homogeneous machines is a problem that has been extensively studied. For heterogeneous machines, with a diverse CPU capacity, new challenges arise, namely the selection of the set of CPUs that optimizes the processing time of a given task. This paper presents a methodology to determine the best allotment for each task that minimizes its processing time on a given heterogeneous machine. The aim is to improve the global processing time of a complex algorithm composed by several linear algebra kernels, by scheduling a number of parallel tasks in the heterogeneous machine. The results presented compare this approach to the simpler data parallel execution of each task on the heterogeneous cluster, i.e. executing one task at a time using the data parallel programming model.

Language: English

Type (Professor's evaluation): Scientific

No. of pages: 6