Analysis of a Self-Organizing Algorithm for Energy Saving in Data Centers

The Cloud computing paradigm allows users to satisfy their increasing need for on-demand and remote computational services. These services are provided by data centers that often consume a huge amount of electrical power. Recently the ecoCloud algorithm has been proposed as a solution for saving energy by consolidating Virtual Machines on as few servers as possible, so as to hibernate the remaining servers and save energy. The ecoCloud approach founds on probabilistic processes: mapping and migration of VMs are driven by Bernoulli trials whose success probability depends on the utilization of single servers. These processes are self-organizing and decentralized, which makes them particularly efficient in large data centers. While in previous work the performance evaluation of ecoCloud was based on artificial traces, in this paper, a mathematical analysis is presented along with simulations fed with logs of real VMs. Results show that efficiency is very close to the theoretical minimum and comparable to that of one of the best centralized algorithms devised so far; in addition, ecoCloud notably reduces the frequency of events, such as VM migrations and server switches, that can deteriorate the quality of service