Predictive Thermal Control for Real-Time Video Decoding

Multimedia applications with soft real-time constraints consume considerable power, incurring thermal problems. It is challenging to process multimedia data in real-time with the highest possible QoS, while avoiding potential thermal problems. To aggravate the problem, immediate dissipation of the accumulated heat is infeasible. In this paper, we design an empirical model to predict the power consumption and resulting processor temperature for video decoding. Based on the model, we develop a predictive method that avoids deadline misses due to thermal faults by adapting the video quality before the temperature exceeds the specified threshold. We implement our predictive control scheme and two baseline approaches by extending an open source implementation of the H.264 scalable video coding standard. Our approach controls the temperature to be below the specified threshold for most of the time unlike the tested baselines, while supporting similar or even better video quality.