Dynamic Virtual Machine Allocation Policy for Load Balancing using Principal Component Analysis and Clustering Technique in Cloud Computing

Abstract

The scalability and agility characteristics of cloud computing allow load balancing to reroute workload requests easily and to enhance overall accessibility. One of the most important services for cloud computing is Infrastructure as a Service (IaaS). There is a large number of physical hosts in a cloud data center for IaaS and it is quite difficult to arrange the allocation of the workload requests manually. Therefore, different load balancing methods have been proposed by researchers to avoid overloaded physical hosts in the cloud data center. However, fewer works have used multivariate analysis in cloud computing environment for considering the dynamic changes of the computing resources. Thus, this work suggests a new Virtual Machine (VM) allocation policy for load balancing by using a multivariate technique, Principal Component Analysis (PCA), and clustering technique. Moreover, PCA and clustering techniques were simulated on a cloud computing simulator, CloudSim. In the proposed allocation policy, a group of VMs were dynamically allocated to physical hosts. The allocation was based on the clusters of hosts according to their similar features in computing resources. The clusters were formed using PCA and a clustering technique based on variables related to the physical hosts such as Million Instructions Per Second (MIPS), Random Access Memory (RAM), bandwidth and storage. The results show that the completion time for all tasks has decreased, and the resource utilization has increased. This will optimize the performance of cloud data centers by effectively utilizing the available resources.