On thermal utilization of periodic task sets in uni-processor systems

In this paper we introduce a novel characterization of real-time tasks based on their temperature impact. This characterization is used to analyze the schedulability of periodic real-time tasks in thermally constrained systems. The proposed characterization is important because thermal constraints are becoming increasingly vital due to rapid and increasing rise in power densities of modern architectures. As part of this work, we introduce the concept of “Accumulated Thermal Impact” (ATI), which represents cumulative temperature increase due to execution of a given task. The concept of ATI is then used to determine thermal utilization of a periodic task set, which is somewhat analogous to traditional computation utilization, and perform schedulability analysis. We also propose a speed scaling scheme for minimizing thermal utilization/system temperature. Our results show that thermal utilization of a periodic task set is strongly correlated to its thermal feasibility. The proposed speed scaling scheme is also shown to perform significantly better than current schemes in terms of thermal utilization/temperature minimization.