Temperature-Aware Speed Scheduling in Periodic Real-Time Systems

Real-time embedded systems usually work in harsh environments with variable ambient temperature. As temperature has a significant impact on the processor reliability, thermal management is mandatory in many such systems, where it is critical to keep the processor temperature below some specified threshold. To achieve this goal, however, the system performance may adversely be affected. Our favorite performance objective in this paper is the minimization of the fraction of jobs which miss their deadline, regarding the thermal constraint. We propose a quasi-static approach based on dynamic voltage scaling: it first generates a number of thermally feasible speed schedules; one schedule for each range of ambient temperatures; then, at runtime, the appropriate speed schedule is selected according to the ambient temperature. An online algorithm is also proposed to decide when the switching between the schedules occurs. The latter algorithm brings the processor temperature below specific values, so the newly selected speed schedule starts to work at the new scheduling without violating the temperature constraint. The efficiencies of the proposed methods are investigated by simulation experiments.