Performance Modelling of Consolidated Virtual Machines

Abstract

With rapid and flexible resource provisioning of virtualization in data centers, problems of determining optimal virtual machine (VM) placements and dealing with virtualization overheads have emerged due to workload fluctuations and changing needs. These challenges have impacts on system performance.In this work, a performance model based on queuing theory, statistical methods and basic theories on system performance is proposed in which the researcher models the response time distribution of an application performance metric conditioned on variables that can be measured or controlled, such as system resource utilization and allocation metrics. The research also examined the relationships between virtualized CPU allocation, CPU contention, and application response time to identify the influence of CPU allocation and how it affects system performance.Comparing estimated values with measured values, empirical result shows that the proposed model validated for all the CPU allocations in the experiments conducted. The response time increases as workload increases; it is also observed from the analysis that the response time increases with low CPU. Thus by varying the CPU allocation base on business needs, am optimal point can be reached such that the CPU can be efficiently managed.