Attacking the Bottlenecks of Back lling Schedulers

Backklling is a simple and eeective way of improving the utilization of space-sharing schedulers. Simple rst-come-rst-served approaches are ineeective because large jobs can fragment the available resources. Backklling schedulers address this problem by allowing jobs to move ahead in the queue, provided that they will not delay subsequent jobs. Previous research has shown that inaccurate estimates of execution times can lead to better backklling schedules. In the rst part of this study, we characterize this eeect on several workloads, and show that average slowdowns can be eeectively reduced by systematically lengthening estimated execution times. Further, we show that the average job slowdown metric can be addressed directly by sorting jobs by increasing execution time. Finally , we modify our sorting scheduler to ensure that incoming jobs can be given hard guarantees. The resulting scheduler guarantees to avoid starvation, and performs signiicantly better than previous backklling schedulers. In the second part of this study, we show how queue randomization and even more a combination of queue randomization and sorting by job length can improve performance. We show that these improvements are better then with queue sorting by job length alone in the simulation with actual estimates of job running times. We investigate the real characteristics of these estimates, and show the wide range of overestimation. To exploit even more randomization and queue sorting, we eliminate guarantees from backklling algorithm, and show signiicant improvements. Finally, we show a limited usefulness of these guarantees, and show that queue sorting criteria can be modiied to prevent starvation in the modiied backklling algorithm.