| Abstract |
Energy-saving in information and communications technology (ICT) systems has become a high priority issue. The efficiency of processor power supplies has seen significant improvements in recent years. However, efficiency under light load conditions needs further improvements since processor loads tend to be light when running practical applications. As a way to improve power efficiency over a wide range of computer workloads, the concept of adaptive power efficiency control, which adaptively interlocks power control with data processing workloads, has been proposed. With this technique, processor power consumption is estimated based on performance monitoring information from the processor without time overhead. However, to produce precise power consumption estimates, appropriate parameters for each application to be executed on the processor are required, and how best to determine the optimal parameters has not yet been established. This study proposes a new online method for determining the most suitable power estimation parameters for each specific application.The proposed method was verified using performance monitoring information and power load data sampled from an experimental computer. The results of the verification show that the method is effective in generating optimal parameters and that it markedly boosts the accuracy of power consumption estimates, thus suggesting its potential for enabling the inclusion of effective adaptive power efficiency controls on general computers. |
