Algorithms of Distributed Power Management for Cloud Computing

N. N. Efanov¹, Junior Software Developer, e-mail: nefanov@parallels.com, A. L. Melekhova^{1, 2}, Lead Software Developer, Lecturer, e-mail: annam@parallels.com, A. O. Bondar¹, Junior Software Developer, e-mail: abondar@parallels.com
¹Parallels, Moscow, ² Moscow Institute of Physics and Technology

Nowadays, when the electricity price is that high, the demand on methods to reduce the power consumption for computing systems is high. While many computational tasks have migrated into the cloud, cloud power consumption becomes a case of study for many researchers. There are lots of investigations in this area, but most of them do not take into account a complete vision of the system. For example, many academic researches ignore the migration price, while most practical solutions are based on empirical results.

In this paper we give the aggregated picture of cloud system from the power management point of view. We cover both global and local power management and make a connection between them. We formulate power management subtasks that should be solved in the cloud system. We discuss the most promising results and do their critical evaluation. We cover dynamic programming solutions, learning tree algorithms, semi-markov representations and many other methods. The key focus is in evaluation of power consumption and power optimization on the local node. In the end we provide our recommendation on distributed power management politics implementation. This work is must-read for all who intend to build their own cloud or develop own cloud solution.

Keywords: power consumption, visualization, cloud computing, dynamic power management, energy proportional, energy efficiency

pp. 20–30