Temperature Minimization and Thermal-Driven Scheduling for Real-Time Periodic Tasks

Abstract

Listen

Translate article

With the increasing power density of processors due to the continuous shrinking of chip size, thermal issue of real-time systems has become more and more urgent since it can greatly affect the systems’ reliability and safety. However, due to the difficulty in quantitative analysis of temperature, existing thermal-aware research either considers temperature as a constraint to minimize energy consumption treating leakage power as a constant, or minimizes the temperature qualitatively through approaches like cool-hot task execution pattern. Even the latest work also only analyzes the average temperature in thermal steady state. To this end, this paper aims at quantitatively minimizing the temperature in both steady and transient states while considering the effect of temperature on leakage power. The following contributions have been made: presents a method of task construction, based on which the minimal temperature condition for any scheduling algorithm like GPS (Global Processor Sharing) and EDF (Earliest Deadline First) under both steady and transient states is proved; discovers a thermal law exhibited by tasks’ power consumption, based on which a thermal-aware approximate algorithm of GPS with a relatively longer scheduling length is designed to mitigate the higher switching overhead of GPS scheduling while guaranteeing the tasks’ timing constraints. Sufficient experiments validate the minimal temperature condition and thermal-aware scheduling algorithm. The two important conclusions obtained from this work are: (1) when the optimal condition can be achieved, all of the scheduling algorithms like GPS and EDF result in the same minimal temperature traces since all of the tasks share the same power consumption value and the processor is fully utilized; (2) when not, the executing sequence of tasks in a scheduling interval of the approximate GPS scheduling, which is determined based on the thermal law, exhibits a better thermal-aware feature.