Efficient Scheduling Heuristics for GridRPC Systems - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Communication Dans Un Congrès Année : 2004

Efficient Scheduling Heuristics for GridRPC Systems

Yves Caniou
  • Fonction : Auteur
  • PersonId : 830502
  • IdRef : 084594896
Emmanuel Jeannot

Résumé

In this paper we examine several scheduling heuristics for GridRPC middleware relying on the time-shared model (a server can execute more than one task at a time). Our work is based on a forecast module called the ‘Historical Trace Manager' (HTM), which is able to predict durations of tasks in the system. We show that the predictions performed by the HTM are very accurate. The five proposed scheduling heuristics use these predictions to map submitted tasks to servers. Experimental simulation results show that they are able to outperform the well-known MCT heuristic for several metrics (makespan but also sumflow, max-stretch, etc.) and therefore provide a better quality of service for the client.

Domaines

Autre [cs.OH]
Fichier non déposé

Dates et versions

inria-00100130 , version 1 (26-09-2006)

Identifiants

  • HAL Id : inria-00100130 , version 1

Citer

Yves Caniou, Emmanuel Jeannot. Efficient Scheduling Heuristics for GridRPC Systems. QOS and Dynamic System Workshop of ICPADS'2004 - 10th International Conference on Parallel and Distributed Systems, 2004, New-Port Beach, California/USA, pp.621--630. ⟨inria-00100130⟩
109 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More