Dynamic real-time scheduling with task migration for handling bag-of-tasks applications on clusters

The scheduling of real-time tasks on clusters is a critical issue for offering quality-of-service (QoS) assurance. A significant challenge in real-time task scheduling is to support various types of applications. In this paper we focus on the scheduling of bag-of-tasks (BoT) applications consisting of many independent tasks. BoT applications are a typically “embarrassingly parallel” type of applications which widely exist in various fields in science and engineering. We propose a dynamic (online) real-time scheduling algorithm referred to as scheduling algorithm with migration (SAM) for handling real-time BoT applications on cluster systems. SAM schedules tasks to the minimum number of processors so that computation power can be saved for unscheduled large tasks. SAM also utilizes task migration to optimize load balancing without undermining the schedulability of the tasks. Performance evaluation results are presented to demonstrate the effectiveness and competitiveness of our approaches when compared to existing algorithms.