Performance Impact of Advance Reservations from the Grid on Backfill Algorithms

Advance reservations are important to guarantee the quality of services of grid jobs. However, the influence of AR from the grid on local parallel schedulers is not clearly known. EASY backfill and conservative backfill are two important parallel scheduling algorithms. This paper examined their procedures to support AR, evaluated their performances impacted from AR by using discrete event simulations with real trace from the parallel workload archive. Results show that as AR request increases, EASY backfill is preferable with higher utilization and lower average expansion factor and average weighted response time for both of local and AR grid jobs. The use of conservative backfill typically provides both of local jobs and AR grid jobs improved predictability and certainty, at the cost of lower utilization and degraded performance for both of local and AR grid jobs.