Please use this identifier to cite or link to this item:
Title: Prediction based dynamic resource provisioning in virtualized environments
Authors: Raghunath, B.R.
Annappa, B.
Issue Date: 2017
Citation: 2017 IEEE International Conference on Consumer Electronics, ICCE 2017, 2017, Vol., , pp.100-105
Abstract: Dynamic provisioning to virtual machines (VMs) is one of the important requirements in the virtualized data centers to make effective utilization of resources. This can be achieved by vertical scaling or horizontal scaling of attached resources. Live virtual machine migration of virtual machines across physical machines is a vertical scaling technique which facilitates resource hotspot mitigation, server consolidation, load balancing and system level maintenance. As live migration is triggered during heavy workload (hotspot) and its procedure takes significant amount of resources to iteratively copy memory pages from source to destination, it affects the performance of other running VMs hosted on the source as well as destination physical machine (PM). Hence to avoid such performance interference effects it is necessary to trigger the migration procedure at such a point where sufficient amount of resources will be available to all the running VMs and to the migrating procedure. It is also important to select such a VM which will produce less performance interference at the source and destination. This paper presents an intelligent decision maker to trigger the migration in such a way that it avoids the said performance interference effects. It predicts the future workload for early detection of overloads and accordingly triggers the migration procedure. It also models the migration procedure to calculate performance parameters and interference parameters which are used in the decision of selection of a VM. Experimental results show that it is able to increase the performance by 45%-50% for network intensive workloads and 25%-30% for CPU, memory intensive workloads when compared with traditional method. It improves the performance by 35%-40% for network intensive workloads and 15%-20% for CPU, memory intensive workloads when compared with Sandpiper method. � 2017 IEEE.
Appears in Collections:2. Conference Papers

Files in This Item:
There are no files associated with this item.

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.