Dynamic speed scaling minimizing expected energy consumption for real-time tasks

摘要

This paper proposes a Markov Decision Process (MDP) approach to compute theoptimal on-line speed scaling policy to minimize the energy consumption of a processor executinga finite or infinite set of jobs with real-time constraints. The policy is computed off-line but usedon-line. We provide several qualitative properties of the optimal policy: monotonicity with respectto the jobs parameters, comparison with on-line deterministic algorithms. Numerical experimentsshow that our proposition performs well when compared with off-line optimal solutions and outperformson-line solutions oblivious to statistical information on the jobs. Several extensions arealso explained when speed changes as well as context switch costs are taken into account. Nonconvexpower functions are also taken into account to model leakage. Finally, state space reductionusing a coarser discretization is presented to deal with the curse of dimensionality of the MDP.