Policy optimization for dynamic power management

Dynamic power management schemes (also called policies) reduce the power co nsumption of complex electronic systems by trading off performance for powe r in a controlled fashion, taking system workload into account. In a power- managed system it is possible to set components into different states, each characterized by performance and power consumption levels. The main functi on of a power management policy is to decide when to perform component stat e transitions and which transition should be performed, depending on system history, workload, and performance constraints. In the past, power management policies have been formulated heuristically, The main contribution of this paper Is to introduce a finite-state, abstrac t system model for power-managed systems based on Markov decision processes . Under this model, the problem of finding policies that optimally tradeoff performance for power can be cast as a stochastic optimization problem and solved exactly and efficiently, The applicability and generality of the ap proach are assessed by formulating Markov model and optimizing power manage ment policies for several systems.