Efficient Scheduling of Real-Time Parallel Applications on Clusters

This paper proposes a novel approach to schedule real-time applications represented by Fork-Join (FJ) task graphs on a cluster platform. The novelty of our approach lies in its ability to efficiently utilize the power of the cluster's computational resources to improve the system throughput. Our approach integrates two heuristic scheduling algorithms: the first algorithm (partition algorithm) works on the cluster level to search for the best allocation scheme for the application's tasks on the cluster's processors. This search is guided by an objective function that aims to optimize the utilization of cluster's resources. The second algorithm (local scheduler) works on the individual processor level to efficiently utilize the processing power of each processor. A set of simulation experiments have been conducted to evaluate the performance of our scheduling approach on both homogeneous and heterogeneous clusters. The results show that our approach improves the acceptance rate of the parallel applications on the cluster compared to traditional approach.