Peak Temperature Minimization for Embedded Systems with DVS Transition Overhead Consideration

Peak temperature is a critical issue in chip multiprocessor (CMP) architecture design. Dynamic Voltage Scaling (DVS) is used nowadays to reduce power consumption and peak temperature. However, traditional DVS in literature usually fails to take into consideration of task dependencies and voltage transition overhead such as energy consumption. In this paper, we propose the ROtation Scheduling and voltage Assignment (ROSA) algorithm, to minimize peak temperature under energy consumption constraints in voltage transition. We conduct experiments on Kernel Linux equipped with 4 Intel Core Quad CPUs. The experimental results demonstrate that our algorithm can reduce the peak temperature on chip with 4.9°C more comparing with the List Scheduling Solution (LSS).