Auto-scaling of Cloud Resources to Manage Surges in Web Application Traffic

Abstract

Cloud computing has become a powerful computing platform for web applications because of the ability to auto-scale based on the demand from the users. Many cloud service providers including Amazon, Azure etc. and software such as RightScale and Scalr provide auto-scaling mechanisms. The current auto-scaling mechanisms seem promising when the application workload follows a certain pattern, but they do not perform well when a sudden surge in traffic hits the web application. They lack the intelligence to learn about the web applications they host. The result of this is often unavailability of the application or very poor performance when underlying application experiences an unpredictable sudden surge. Most of the existing techniques of auto-scaling are based on simple user defined metrics and resource utilization thresholds. These approaches mainly focus on the static allocation and do not work well with the present dynamic natured web applications where work load is highly unpredictable. This research work proposes a solution for autonomous dynamic scaling and reconfiguring clouds when web applications suffer a sudden large and unpredictable swing in traffic. It makes use of the intelligence of the multi-agents to detect the application behavior for unusual traffic. Machine learning techniques are applied to accurately predict the future workload and a surge detection mechanism to look for potential surges. The proposed system also contains a planner agent which carefully computes the resource requirement to handle the incoming traffic surge. We have proposed the use of a resource buffer as quick solution for the sudden surges in traffic which reduces the VM’s churn time and makes sure timely availability of the resources when resource demand increases suddenly due to the surge in traffic.