A Robust Decentralized Job Scheduling Approach for Mobile Peers in Ad-hoc Grids

Abstract

The increasing capabilities and spreading of mobile technology raise the opportunity to integrate potentially unstable mobile devices as resources into grids. In this work, we contribute by proposing and evaluating a robust decentralized job scheduling approach for mobile peers forming an ad-hoc grid. The scheduling approach is based on a first come first serve strategy executed locally by each peer. Additionally, the peer performs matchmaking between a job's requirements and the device's capabilities autonomously. Coordination between mobile peers is based on job queues shared within a distributed virtual shared memory (VSM). By applying proactive and reactive fault tolerance mechanisms we were able to increase the robustness of the scheduler. A prototypical implementation using the VSM CORSO, Java, and Condor ClassAds demonstrates the feasibility of our approach. The conducted experiments show that the job loss ratio and the average job response time can be decreased by adding fault tolerance mechanisms.