A distributed control framework for performance management of virtualized computing environments

Abstract

This paper develops a distributed cooperative control framework to manage the performance of virtualized computing environments. We consider a server cluster hosting multiple enterprise applications on a set of virtual machines (VMs) in which the system must dynamically optimize the CPU capacity provided to each VM in response to incoming workload intensity such that desired response times are satisfied. We solve the overall control/optimization problem by decomposing it into a set of smaller subproblems that can be solved cooperatively by individual controllers. Model-predictive controllers, implemented locally within each server, independently decide the CPU capacity to allocate to VMs under their control such that the overall system's performance goals are satisfied. We experimentally validate the proposed framework on a server cluster supporting three online services, showing that our scheme is highly scalable, naturally tolerates server failures, and allows for the dynamic addition/removal of servers during system operation without requiring changes to the overall control architecture. © 2010 ACM.