Workload Prediction Model for Autonomic Scaling of Cloud Resources with Machine Learning
Sanjay T. Singh, Mahendra Tiwari, Anchit Sajal Dhar · 2023
Cloud computing enables clients with on-demand access to software, platform, and infrastructure, in the form of services through the Internet. Client applications are executed over the cloud which is backed by Virtual Machines (VMs) and these VMs are hosted on top of physical servers. The amount of workload traffic received by the cloud changes over time. To fulfil these fluctuating workload needs, VMs must be automatically scaled up and down to guarantee that the quality of service (QoS) to the client is maintained, which, in turn, must be achieved by ensuring that the Service-Level Agreement (SLA) criteria are not breached. To achieve this goal of automatic scaling (also known as auto-scaling), the important task is to predict the future workload demands for cloud resources so that appropriate numbers of VMs must be made ready in advance so that the requirements of clients are met. The prediction is done on the basis of the past resource usage trends. In this paper, we propose an autonomic 344 resource management model, based on a Machine Learning (ML) technique called Random Forest, that attempts to solve the problem of predicting the cloud resource demands. We have evaluated the proposed classifier, for its prediction accuracy by utilizing an actual datacenter dataset containing thousands of records. Further, the classifier is compared with another classifier based on Naïve Bayes algorithm and found that our proposed algorithm gives better performance. The results obtained indicate the accuracy and efficiency of Random Forest Classifier.