A Formal Approach to Frequent Energy Adaptations for Multimedia Applications

Much research has recently been done on adapting architecturalresources of general-purpose processors to saveenergy at the cost of increased execution time. This workexamines adaptation control algorithms for such processorsrunning real-time multimedia applications. The best previousalgorithms are mostly heuristics-based and ad hoc, requiringan impractically large amount of application- andresource-specific tuning.We take a more formal approach that does not require thelarge tuning effort of previous approaches, and yet obtainsaverage energy savings comparable to the best previous approach.We pose control algorithm design as a constrainedoptimization problem: what configuration should be used ateach point in the program to minimize energy for a targetedperformance given that each configuration has a differentenergy-performance tradeoff at each point? We solve thiswith the method of Lagrange multipliers, which assumesknowledge of the energy-performance tradeoffs. We developa technique to estimate these tradeoffs using properties ofmultimedia applications. Our technique is likely extendibleto other application domains.We compare our algorithm to the best previous algorithmfor real-time multimedia applications, which is heuristics-based.We demonstrate the practical difficulty of the tuningprocess for the previous algorithm. Compared to apainstakingly hand-tuned version of that algorithm, ournew algorithm provides similar energy savings through amore formal approach that does not need such heroic tuning,making it practical to implement.