Necessary and Sufficient Conditions for Thermal Schedulability of Periodic Real-Time Tasks Under Fluid Scheduling Model

With the growing need to address the thermal issues in modern processing platforms, various performance throttling schemes have been proposed in literature (DVFS, clock gating, and so on) to manage temperature. In real-time systems, such methods are often unacceptable, as they can result in potentially catastrophic deadline misses. As a result, real-time scheduling research has recently focused on developing algorithms that meet the compute deadline while satisfying power and thermal constraints. Basic bounds that can determine if a set of tasks can be scheduled or not were established in the 1970s based on computation utilization. Similar results for thermal bounds have not been forthcoming. In this article, we address the problem of thermal constraint schedulability of tasks and derive necessary and sufficient conditions for thermal feasibility of periodic tasksets on a unicore system. We prove that a GPS-inspired fluid scheduling scheme is thermally optimal when context switch/preemption overhead is ignored. Extension of sufficient conditions to a nonfluid model is still an open problem. We also extend some of the results to a multicore processing environment. We demonstrate the efficacy of our results through extensive simulations. We also evaluate the proposed concepts on a hardware testbed.