Dynamic Mapping of Application Workflows in Heterogeneous Computing Environments

Abstract

Performance of a Heterogeneous Computing Environment (HCE) mainly depends on the efficiency of application workflow scheduling algorithms. Achieving high efficiency of application workflow scheduling algorithms in HCE is an NPComplete problem. A novel application workflow scheduling algorithm called Heterogeneous Dynamic List Task Scheduling (HDLTS) for HCE is proposed in this paper. The functionality of HDLTS majorly relies on the following three pillars; first, duplicate the entry task only if it helps to reduce the overall application execution time; second, for mapping, consider only those tasks that have all the necessary input conditions to start the execution and find out the heterogeneity of their execution time on the computational resources; third, select the task that has higher execution time heterogeneity, and map it to a resource that takes minimum time to execute the task. The HDLTS task selection and mapping policies dynamically consider the resource utilization and task assignment that makes it more efficient and enables it to produce good quality schedules. The performance of the HDLTS is evaluated against popular list scheduling algorithms on randomly generated application workflows and real world application workflows. Experimental results prove that the HDLTS outperforms well-known list scheduling algorithms, such as in terms of schedule length and efficiency.