Fairness Evaluation of Backfilling Schedulers using the Net-Benefit Approach

John Ngubiri


Fairness, Scheduler, net-benefit, backfilling


Fairness is a very important aspect in queuing systems. Possibly it is because of fairness that queues came into existence in the first place. Many fairness metrics have been proposed for parallel job schedulers. Likewise, many studies have been made analyzing their suitability in evaluating fairness. One of the approaches proposed is the net-benefit approach. However, less studies have been made to evaluate its suitability and consistence in evaluating fairness. In this paper, we analyze the approach using the backfilling schedulers. We use First Come First Serve as a base scheduler. We show that aggressive backfilling is fairer that conservative backfilling. We also show that look ahead technique improves the fairness of the aggressive and conservative backfilling but does not affect their relative fairness. Finally we show that at a higher load, the difference in fairness of conservative and aggressive backfilling gets wider.

Important Links:

Go Back