Efficient DAG Scheduling with Resource-Aware Clustering for Heterogeneous Systems

Task scheduling on Heterogeneous Distributed Computing Systems (HeDCSs) with the purpose of efficiency and reduction of execution time is of paramount importance. In this paper a novel task scheduling algorithm, called Resource-Aware Clustering (RAC) for Directed Acyclic Graphs (DAGs) is proposed. The objective of this algorithm is to keep the relative load balancing and efficiency increase between processors with different processor capabilities. To aim this fact, RAC by giving a dynamic score function to each task, performs task clustering and allocation according to processing capability of cooperative processors. In execution phase, Modified Bottom-Level (MBL) quantity of each task with a complexity of O(v+e) would be calculated and applied for tasks priority identification. Comparing simulation results of RAC algorithm with famous scheduling approaches such as MCP, MD and DSC shows that RAC substantially increases efficiency while in most cases reduces completion time.