Dynamic frequency scaling algorithms for improving the CPU´s energy efficiency

This paper approaches the problem of improving the service center server CPU´s energy efficiency by executing dynamic frequency scaling actions and performing tradeoffs between CPU´s computational performance and its power consumption. Two different algorithms are designed and implemented: an immune inspired algorithm and a fuzzy logic based algorithm. The immune inspired algorithm uses the human antigen as a model to represent the server power / performance state. Using a set of detectors the antigens are classified as self for optimal power consumption state or non-self for non-optimal power consumption state. For the non-self antigens a biologically inspired clonal selection approach is used to determine the actions that need to be executed to bring the server´s CPU in an optimal power consumption state. The fuzzy logic based algorithm adaptively changes the processor performance states to the incoming workload. The algorithm also filters workload spikes because frequent p-states transition costs can outweigh the benefit of adaptation.