Decentralized Preemptive Scheduling Across Heterogeneous Multi-core Grid Resources

Abstract

The recent advent of multi-core computing environments increases the heterogeneity of grid resources and the complexity of managing them, making efficient load balancing challenging. In an environment where jobs are submitted regularly into a grid which is already executing several jobs, it becomes important to provide low job turnaround times and high throughput for the users. Typically, the grids employ a First Come First Serve (FCFS) method of executing the jobs in the queue which results in suboptimal turn-around times and wait times for most jobs. Hence a conventional FCFS scheduling strategy does not suffice to reduce the average wait times across all jobs. In this paper, we propose new decentralized preemptive scheduling strategies that backfill jobs locally and dynamically migrate waiting jobs across nodes to leverage residual resources, while guaranteeing (on a best effort basis) bounded turn-around and waiting times for all jobs. The methods attempt to maximize total throughput and minimize average waiting time while balancing load across available grid resources. Experimental results for both intra-node and internode scheduling via simulation show that our scheduling schemes perform considerably better than the conventional FCFS approach of a distributed or a centralized scheduler.

DOI: 10.1007/978-3-662-43779-7_2

Extracted Key Phrases

6 Figures and Tables

Cite this paper

@inproceedings{Balasubramanian2013DecentralizedPS, title={Decentralized Preemptive Scheduling Across Heterogeneous Multi-core Grid Resources}, author={Arun Balasubramanian and Alan Sussman and Norman M. Sadeh}, booktitle={JSSPP}, year={2013} }