PREDICTIVE PROVISIONING OF WORKLOADS FOR DYNAMIC APPLICATION SCALING IN CLOUD ENVIRONMENTS

The large scale emergence of cloud platforms induce the tendency to virtualize application workloads that traditionally ran on physical machines. At the same time, cloud providers advertise unlimited resources available to the customers at any time for a fixed price. These factors create the opportunity for customers to easily scale up and down the infrastructure depending on the real time requirements, reducing the overall costs for providing the service. Cloud platforms today provide a threshold trigger mechanism that can trigger provisioning or deprovisioning of additional resources. This paper argues that the threshold approach is not enough for some real life application scaling requirements and introduces a predictive mechanism that allows accurate and proactive provisioning of workloads. The prediction algorithm is based on the observation that for some applications a usage pattern exists, and this usage pattern is repetitive. This paper presents the usage pattern identified in a large scale travel booking application and the execution of the algorithm on this data. The algorithm tested using IBM CloudBurst 2.1 deployment using a benchmark application and results are discussed.