On-Demand VM Placement on Cloud Infrastructure

Abstract

Cloud Computing paradigm is most popular because of its exibility for provisioning resources quickly and eciently. In cloud computing the resource requests are served by creating virtual machines of the requested specication on the underlying physical infrastructure. If the placement of virtual machines to the underlying physical machines will take long time or if all the accepted virtual machine requests can't be served then some exibility will lost. In on-demand access to cloud computing services the requested resources are served on the available infrastructure for short span of time. In on-demand access the number of resource requests in a particular time interval can not be predicted unlike in case of spot-market access. As a virtual machine instance will run on a single physical machine at a time, hence to serve more requests in case of on-demand access we have to use the available resource optimally considering the allocation cost and SLA violation. In this work we tried to improve the resource utilization by considering single dimensional best strategy, which not only reduce the cost by utilizing minimum number of resources but also minimize the SLA violation which may arise due to failure in allocating all the requested virtual machine. We have developed a framework that optimizes the use of physical infrastructure by eectively allocating the requested virtual machines and also reduces the allocation time. The proposed allocation policy is compared with three other existing policies named Greedy First Fit, Ranking and Round-Robin, by simulating all policies using CloudSim toolkit and the performance is evaluated by considering various parameters.