Jun 29, 2011
Time Series based Dynamic Frequency Scaling Solution for Optimizing the CPU Energy Consumption
Tudor Cioara, Ionut Anghel, Ioan Salomie, Georgiana Copil, Daniel Moldovan, Marius Grindean, Technical University of Cluj-Napoca, Cluj-Napoca, Romania, Time Series based Dynamic Frequency Scaling Solution for Optimizing the CPU Energy Consumption, IEEE 7th International Conference on Intelligent Computer Communication and Processing, August 25-27, 2011, Cluj-Napoca, Romania, http://www.iccp.ro/iccp2011, Special Session: Green Computing, Organized in cooperation with GAMES EU FP7 project
Posted by: alerox
In this paper the problem of service center servers high energy consumption is tackled by proposing a time series based CPU dynamic frequency scaling algorithm. The algorithm senses the workload changes and adapts the CPU power states while minimizing the CPU energy consumption. Our solution analyzes the CPU workload time series for identifying the frequent workload patterns. For each frequent pattern, the corresponding energy efficient dynamic frequency scaling actions are determined and associated using information about the pattern’s sub-sequences trends. A workload characterization function is defined and used to identify the pattern trends. To identify the membership of the new CPU workload observations to a frequent CPU workload pattern, a sliding window based method is used. If such a match is found, the dynamic frequency scaling actions associated to the frequent pattern are executed and the pattern occurrence probability is increased.