Multi-Site Allocation Policies on a Grid and Local Level

Abstract

Efficient job scheduling in computational grids is a challenging task, especially when the workload consists of jobs submitted in a grid and local level. In this study, we consider such a grid system where both local jobs and grid jobs require service. The goal is to maintain a balance between the two competitive job types, in order for every job to be executed in a timely manner. However, local jobs are of higher importance compared to the grid jobs and it is imperative that their waiting time be minimized. Grid jobs are parallel jobs so gang scheduling is implemented, along with various other scheduling techniques in order to improve performance, such as backfilling. A simulation model is considered to evaluate system performance, and experiments are conducted to determine which proposed scheduling policy provides the best results.