Adaptive Virtual Machine Migration Mechanism for Energy Efficiency

Cloud systems have become a popular platform for businessapplications due to the exibility in usage and payment theyoer. One of the caveats of Cloud systems is their high en-ergy consumption. Minimizing energy consumption whilemaintaining a high service level has become a relevant opti-mization task for Cloud providers. Opportunities for energysavings arise when server hosts are overloaded, which alsoentails unnecessary delays.To address the problem, researchers have devised strate-gies how to choose the server host to deploy an application toand how to choose a running application for migration whena host has been identied as overloaded. In this work, weintroduce a Bayesian Belief Network which learns over timewhich of the virtual machines are best removed from a hostthat has been identied as overloaded. The probabilisticchoice is made among virtual machines that are grouped bytheir degree of CPU usage. Given the feedback in the formof the computing resources saved, the system learns whichvirtual machine proles should be shifted for best perfor-mance. This strategy compares favourably to two existingmethods for load balancing.