Analyzing Fuzzy Models for VM Resource Usage Prediction in Virtualized Data Center

Abstract

Data Centre (DC) administrators try to delivering performance guarantees while managing resources for utilization in terms of reducing cost. With the advent of server consolidation provided by virtualization technology, multiple heterogeneous virtual machines (VMs) can be coexisted on a physical server and shared resources together. Fixed allocation of resources to VMs is not the optimal allocation method as over provisioning and under provisioning can be caused. For dynamic allocation, simultaneous on demand provisioning of shared physical resources to VMs becomes the key challenge. This paper proposes resource usage prediction system by making analysis on the accuracy of three different models; Fuzzy modeling, adaptive Fuzzy modeling and NeuroFuzzy modeling. To evaluate the efficiency of three different models, experiments are carried out by workload-resource mapping and resource-resource mapping approaches. CPU profiles from real world data centre are used to analyze through a simulating program. Experimental results show that the proposed resource prediction models can predict well next time interval resource usage of virtual machine even in the condition of unexpected high spikes CPU.