Efficient Task Partitioning and Scheduling for Thermal Management in Multicore Processors

Power and heat density of integrated circuits (ICs) are rising exponentially over the years. The overheating of ICs leads to higher cost of cooling and packaging as well as reliability concerns and shorter lifetime. While existing task-partitioning based approaches are promising for reducing peak temperature in uniprocessor systems, there are no previous efforts in exploring temperature-aware task partitioning in multicore architectures. In this paper, we propose a task partitioning and scheduling algorithm to reduce the hot-spot in multicore embedded systems running a set of independent tasks. Experimental results using real benchmarks show that our approach is able to reduce the peak temperature by as much as 4.52C compared to the stateof-the-art thermal-aware task scheduling algorithm PDTM [22] while requires 31% less time to finish all the tasks.