Temperature Aware Scheduling for Embedded Processors

Abstract

Decreasing feature sizes and increased emphasis on performance have resulted in an exponential increase in power density. Higher power density results in high on-chip temperature, which in turn is pushing temperature management to the forefront of computer systems design. The problem of thermal management is more severe in embedded systems as the small size and mobility requirements do not allow for complex cooling solutions. Also traditional hardware based thermal management schemes are not suitable for embedded systems because they are complex and do not account for system level requirements such as real time constraints, fairness, etc. In this paper, we propose a system level thermal management scheme for embedded systems. We first characterize the thermal behavior of a set of embedded tasks executing on an embedded processor and observe that there are large variations in the thermal behavior of different tasks. Our thermal management strategy exploits this observation and manages temperature by controlling the amount of processor time provided to the hot and cold tasks. Using our thermal management strategy, we design a temperature aware scheduler that employs task scheduling and voltage scaling to satisfy soft real time constraints and fairness while maintaining the temperature below a threshold.