Workload prediction and weighted rule-based task scheduling for face certification system on distributed parallel computing

초록

This paper presents a workload prediction and weighted rule-based task scheduling for face certification on distributed parallel computing. To compose a large-scale certification system, such as a criminal surveillance system for a public security, the system requires an enormous processing power. Thus a grid and distributed parallel computing is an essential approach for a large scale certification system. However his kind of approach is generally comprised of heterogeneous resources. And differential characteristics of each resource have influence on a performance of system. Therefore, an efficient task distribution and scheduling is necessary to improve a performance of system. There are various kinds of scheduling for task distribution. However existing methods cannot provide a suitable task distribution for a face certification system. Therefore, this paper proposes a task scheduling which includes a queue management policy with workload-prediction and weighted rule-based resource selection. The proposed method predicts the volume of certification task for a task queue management policy and selects the suitable certification server using performance weighted rules. Simulation result shows that the proposed method has better performance than other scheduling methods. © 2011 Springer-Verlag.