Static Thermal-Aware Task Assignment and Scheduling for Makespan Minimization in Heterogeneous Real-Time MPSoCs

Abstract

In this paper, the authors address the problem of allocating and scheduling tasks of bag-of-tasks applications (BoTs) to multiprocessors for achieving makespan minimization under the thermal and timing constraints. The proposed scheme first selects the processor with highest allocation probability for every task. The allocation probability is calculated under the consideration of processor workload and temperature profiles. In addition, the higher allocation probability of a processor indicates the better performance in terms of makespan and temperature can be achieved by executing the task on this processor. Then, the operating frequencies of tasks are determined and tasks on the processor are executed in the alternate order of being hot-cool to reduce the on-chip peak temperature. Task splitting, that is, splitting a hot task into multiple sections and executing the hot subtasks with idle time alternatively, is also utilized to ensure the peak temperature constraint. Extensive simulations were performed to validate the effectiveness of the proposed approach. The proposed scheme achieves 15.31% and 19.56% reduction in makespan as compared to benchmarking scheme RATM and ?-VSTM, respectively. The peak temperature of the proposed algorithms can be up to 4.38% and 4.49% lower than that of benchmarking schemes, respectively.