Thermal constrained workload distribution for maximizing throughput on multi-core processors

Power density and heat density of multi-core processor system are increasing exponentially based on Moore´s Law. The reliability of a chip is severely impacted by temperature “hot spots”. In this paper, we study scheduling algorithms for multi-core processors that incorporate temperature constraints. Our goal is to optimize the workload distribution on a multicore processor to maximize throughput with a given maximum operating temperature. Our algorithms are targeted for a larger class of data parallel and task parallel applications. Experimental results show that our algorithms are computationally fast, maximize throughput and provide effective temperature management.