Performance Evaluation of Adaptive Scheduling Algorithm for Shared Heterogeneous Cluster Systems
Cluster computing systems have recently generated enormous interest for providing easily scalable and cost-effective parallel computing solution for processing large-scale applications. Various adaptive space-sharing scheduling algorithms have been proposed to improve the performance of dedicated and homogeneous clusters. But commodity clusters are naturally non-dedicated and tend to be heterogeneous over the time as cluster hardware is usually upgraded and new fast machines are also added to improve cluster performance. The existing adaptive policies for dedicated homogeneous and heterogeneous parallel systems are not suitable for such conditions. Most of the existing adaptive policies assume a priori knowledge of certain job characteristics to take scheduling decisions. However such information is not readily available without incurring great cost. This paper fills these gaps by designing robust and effective space-sharing scheduling algorithm for non-dedicated heterogeneous cluster systems, assuming no job characteristics to reduce mean job response time. Evaluation results show that the proposed algorithm provide substantial improvement over existing algorithms at moderate to high system utilizations.
