Adaptive Group-Based Job Scheduling for High Performance and Reliable Volunteer Computing

This paper presents an adaptive group-based job scheduling method for credibility-based sabotage-tolerance techniques in volunteer computing (VC) systems. Credibility-based technique is a promising approach for reliable VC systems since it guarantees computational correctness mathematically based on the credibility of participants. Check-by-voting reduces the cost of checking credibility in credibility-based technique. However, in some applications where the deadline of the computation is relatively short, current job scheduling methods do not work well for check-by-voting and significantly degrade performance. To improve the performance of VCs, the proposed job scheduling method adaptively groups participants based on the expected-credibility to take into account the participants under job execution. Simulation of VCs shows that the proposed method always outperforms current job scheduling methods regardless of the values of unknown parameters such as population and behavior of saboteurs.