Incentives to Tight the Runtime Estimates of EASY Backfilling

Authors:
Li Bo;Jun Chen;Man Yang;Erfei Wang
Affiliations:
School of Information Science and Engineering, Yunnan University, Kunming, China 650091;School of Information Science and Engineering, Yunnan University, Kunming, China 650091;School of Information Science and Engineering, Yunnan University, Kunming, China 650091;School of Information Science and Engineering, Yunnan University, Kunming, China 650091
Venue:
ICDCN '09 Proceedings of the 10th International Conference on Distributed Computing and Networking
Year:
2009

Citing 5
Cited 0

Utilization, Predictability, Workloads, and User Runtime Estimates in Scheduling the IBM SP2 with Backfilling

IEEE Transactions on Parallel and Distributed Systems
Scheduling Jobs on Parallel Systems Using a Relaxed Backfill Strategy

JSSPP '02 Revised Papers from the 8th International Workshop on Job Scheduling Strategies for Parallel Processing
A comprehensive model of the supercomputer workload

WWC '01 Proceedings of the Workload Characterization, 2001. WWC-4. 2001 IEEE International Workshop
Backfilling Using System-Generated Predictions Rather than User Runtime Estimates

IEEE Transactions on Parallel and Distributed Systems
The Impact of Under-Estimated Length of Jobs on EASY-Backfill Scheduling

PDP '08 Proceedings of the 16th Euromicro Conference on Parallel, Distributed and Network-Based Processing (PDP 2008)

Quantified Score

Hi-index	0.00

Visualization

Abstract

In standard backfill algorithms, a job will be terminated immediately it has been processed for its estimated runtime. This drives users not to provide tight estimates to make their jobs to be scheduled earlier but at the risk of being killed. This paper evaluated an enhanced mechanism to tolerate jobs to continue processing if they will not delay any other running or reserved jobs. By comparing the performance of this new mechanism with that of the standard easy backfill by simulation of real workloads, it is shown that both user- and system- centric performance metrics were improved. As more users underestimate the runtime of their jobs and make the estimates tighter, user-centric metrics always be improved, but the system-centric metric in some cases may degrade very slightly.