Local Queueing-Based Data-Driven Task Scheduling for Multicore Systems

Nowadays, multicore systems are widely used in high performance computing. Many algorithms have been proposed to enhance the system performance by load balancing or concurrent scheduling to reduce the execution time of applications. However, task scheduling on multicore systems is still an open issue, which needs to be analyzed to fully utilize the processing capacity and achieve low processing latencies. In order to tackle the inefficient utilization of CPU cores, a queueing-based data-driven task scheduling scheme, which focuses on local parallel computing, is introduced in this paper. In this scheduling scheme, multi-queue management is proposed for dynamic task scheduling to target a full utilization of local CPU cores when input tasks can keep them fully used. Furthermore, the preemption technique is applied to guarantee that high priority tasks will not be blocked by low priority tasks. Our solution can be combined with other algorithms taking into account earliest finish time or critical path to generate better results. Thus CPU core utilization can be improved while minimizing the makespan of high priority DAGs. Finally, simulations are carried out to verify the proposed task scheduling scheme. The reported results confirm its viability and efficiency.