PVBTS: A NOVEL TASK SCHEDULING ALGORITHM FOR HETEROGENEOUS COMPUTING PLATFORMS

Efficient task scheduling has always been one of the most critical issues for high performance in heterogeneous computing. The heterogeneity of computation costs on a given set of processors and the communication costs among processors increase the complexity of the scheduling problem. Generally, the application consists of several tasks with dependencies. If the computation costs, task dependencies and communication costs are known a priori, the application can be represented by a static model, namely the directed acyclic graphs (DAG) model. In this paper, we proposed a novel task scheduling algorithm called penalty value based task scheduling (PVBTS) for application scheduling problem. The PVBTS algorithm dynamically determines the execution order of tasks according to the penalty value which is computed based on the heterogeneity of execution completion time on a given set of processors. In each step, the PVBTS algorithm maintains a ready list including all the independent tasks, then selects the task with the highest penalty value and maps it to a processor that gives the minimum execution completion time of the task. The PVBTS algorithm uses randomly generated task graphs and some real-world application task graphs to evaluate performance. The experimental results indicate that the PVBTS algorithm outperforms some well-known scheduling algorithms selected for the performance comparison in terms of schedule length (makespan) and efficiency.