A Group-Based Job Scheduling Method for Parallel Volunteer Computing

Toward the realization of parallel Volunteer Computing (VC), we propose a group-based job scheduling method based on the expected completion probability. A critical problem that must be addressed in the parallel VC is the volatility of nodes (workers), if any workers leave the VC system, jobs may never be completed due to the inability to communicate with the missing workers. We first define a new parallel VC model and then propose a group-based job scheduling method. We focus on the approach of redundant computation used for removing erroneous results and extend it to deal with the volatility of workers. In the proposed job scheduling method, groups of workers are determined adaptively for each job by calculating expected completion probability of the job considering worker defection rate. This method allows to increase the probability of job's completion, thus leading to the reduction in the computation time. Experimental results indicate that the proposed method reduces completion time of VC about 60%, compared to a simple method which does not consider the worker defection.