TASOR: A Temperature-Aware Semi-Partitioned Real-time Scheduler

Modern multicore systems, which execute complex functionalities on densely packed multi-million gate platforms, are often prone to unacceptable surges in core temperatures. Increase in temperature in such systems not only results in high cooling costs but also leads to high leakage power dissipation, along with reduced efficiency and lower life-span for the system. Given, a set of periodic real-time tasks to be executed on a thermally constrained multicore system, proportional fair schedulers form an attractive scheduling alternative. This is because of their flexibility and the ability to deliver efficient resource utilization, which can potentially enable accurate control over the timeliness of all tasks as well as stipulated temperature upper bounds on all processing cores, over the entire schedule length. In this paper, we propose a low-overhead three-level hierarchical temperature-aware semi-partitioned proportional fair scheduler, called Temperature-Aware Semi-partitioned Real-Time Scheduler (TASOR). The first level in TASOR partitions, time into discrete slices based on task deadlines, such that accurate proportional fairness is maintained at all slice boundaries. In the second level, it classifies each task as either hot or cool, based on it's temperature characteristics. At the last level, TASOR performs intra-slice scheduling with the objective of maximizing resource utilization. Our experimental results show that TASOR is able to perform appreciably under various realistic scenarios.