Dynamic Windows Scheduling for Virtual Machine Placement

Abstract

We consider applying Windows Scheduling (WS) to Virtual Machine(VM) placement in cloud computing. In brief, the problem is to schedule the use of Physical Machines(PM) in terms of mips, processing elements, ram, storage and/or bandwidth to clients. Each client is characterized by an active cycle and a broker which is acting on behalf of that client. The broker hides virtual machine management, as VM creation, submission of cloudlets to VM and destruction of VMs. During the period of time that any given client is active, there must be transmissions from the broker to the cloud resource provider, for example, a Data Center, for VM creation and destruction. The goal of the Data Center Virtual Machine Allocation Policy is to minimize the number of PMs used for VM placement. With this online model, decisions are dynamic, we assume that VMs may be reallocated at some cost. We assume that such cost is a constant amount paid per reallocation. That is, we also aim to minimize the number of reallocations. We will present several online reallocation algorithms for VM placement. We will eval- uate experimentally these policies showing that, in practice, all achieve constant amortized reallocations with close to optimal PM usage. Our simulations will also expose interesting trade-offs between reallocations and PM usage. We are developing the simulation model termed Virtual- MachinePlacementSim (VMPSim) based on CloudSim to perform the experiments.