VM consolidation or placement using utilization prediction model and scheduling algorithms

Abstract

Among the challenges faced by most of the business CIOs and IT managers now-a-days, among them, mostly are profitable or efficient utilization of the IT infrastructure, degree of the responsive-ness in urging for new initiatives of business and is much flexible in acclimatizing to changes in the organisation. The continued worries of meeting the IT budget as well as the stringent regulatory requirements are also the main challenges faced by the IT industry. Virtualisation, a technological innovation, helps a lot in deploying creative solutions to these business challenges. Data centre virtualisation is defined as the process of developing, deploying and designing a data centre on virtualization and other computing technologies. It helps to virtualize physical servers in a data centre and does this along with different aspects like storage devices, networking devices and other infrastructure and equipment. [1] Data centre virtualization usually produces a collocated and virtualized cloud data centre. By research, we know that, consolidation of the VMs helps to optimise or in fact reduce the usage of resources thereby reducing unnecessary energy consumption in a cloud data centre. The VM placement has a vital impact in the consolidation of the VMs. The specialists have created many different calculations or algorithms for VM consolidation considering the efficient energy utilization. However, these algorithms lack the use of exploitation mechanism efficiently and most of them focussed on number of physical machines (PMs) minimization and neglected future resource demands. Due to this, there were unnecessary VM migrations done [2]. Moreover, it violated most of the rules of the Service Level Agreement (also abbreviated as SLA) in data centres. In order to resolve this issue, this paper proposes an approach via VM consolidation that considers both, current as well as future utilization of resources. There are two solutions provide where one uses a model based on regression algorithm [3] in order to approximate and predict the CPU and memory utilizations of VMs and PMs and other uses two separate designed algorithms specifically designed to schedule VM in multi-tenant data centers. In order to attain the analysis and performance part, real workload traces like PlanetLab and Google cluster [4] are used. The results achieved via the proposed solutions showed betterment over the heuristic and meta-heuristic algorithms used here in reducing particularly the consumption of energy, the rate of migrations of VM and also the reduction of number of violations of SLA.