Résumé: Scheduling and resource management play an important role in building complex distributed systems, such as grids. In this article we study the impact on performance of job service demand variability in a two-level grid architecture, given that the grid and local schedulers are unaware of each job’s service demand. We examine two scheduling policies at grid level, which utilize site load information and three policies at local level. A simulation model is used to evaluate performance. Results show that service demand variability degrades performance, and thus a suitable local resource allocation policy is needed to reduce its impact.

Langue: Anglais