Cross-Correlation Prediction of Resource Demand for Virtual Machine Resource Allocation in Clouds
Dorian Minarolli, Bernd Freisleben · 2014
Cloud computing is aimed at offering elastic resource allocation on demand in a pay-as-you-go fashion to cloud consumers. To achieve this goal in automatic manner, a resource scaling mechanism is needed that maintains application performance according to Service Level Agreements (SLA) and reduces resource costs at the same time. In this paper, we present a cross-correlation prediction approach based on machine learning that predicts resource demands of multiple resources of virtual machines running in a cloud infrastructure. Based on these predictions, a proactive resource allocation scheme is applied that assigns only the required resources to virtual machines to keep their cost to a minimum. Experimental results with the web serving multi-tier application benchmark of CloudSuite show the effectiveness of our approach compared to a non-cross-correlation prediction technique in achieving better prediction accuracy and better application performance.