A Novel Dynamic Virtual Machine Consolidation Algorithm Based on Correlation Coefficient Analysis and Bin Packing

Abstract

Virt ual Machine Consolidation (VMC) is known as a crucial method to improve system resource utilization and service level agreement in a datacenter. In this paper, we propose a novel algorithm named Segmentation Iteration Correlation Combination (SICC) expressly for the VMC. In this algorithm we integrate the methods of statistic regression modeling, Pearson correlation coefficient analysis and off-line Bin packing to establish a new process strategy to achieve excellent higher one dimensional resource utilization of a datacenter. The SICC operates based on an innovative two-stage strategy: the first stage is to divide all the Virtual Machines (VM) into several groups and reduce the peak-mean difference value of the VM resource utilization inside each VM group as much as possible by the Correlation Coefficient Serial analysis and one kind of improved VM dynamic complementary consolidation algorithm, derived from the algorithm of Iterative Correlation Match Algorithm (ICMA). When the difference of peak-mean value is small enough, we can take advantage of the Bin Packing methods in the second stage to improve resource utilization on account of the reasonable VM consolidation order. The numerical simulation indicates that the algorithm feature 3 % to 20% performance improvement in resource utilization to ICMA algorithm and approximate 50% performance improvement in resource utilization to First Fit Decreasing (FFD) with the same dynamic initial conditions.