Machine Learning based Workload Prediction for Auto-scaling Cloud Applications

Cloud computing is a ubiquitous computing paradigm that offers its users access to software, platforms, and infrastructure as services, on-demand, over the Internet. User requests for these services (also known as workload) are placed over Virtual Machines (VMs) for execution. These VMs are hosted over Physical Machines (PMs) to abstract processing capabilities of these PMs. Over a period, Cloud services experience fluctuations in the workload pattern. To match the resources required for serving the varying workload, VMs must be added or removed autonomically as performing the same task manually is inefficient. Moreover, VMs take some fraction of time to be setup before they can be used. The goal behind automatic scale-up and scale-down operations is to ensure that the Service Level Agreement (SLA) between the cloud service provider and cloud client is upheld and cloud users experience acceptable levels of Quality of Service (QoS). The essential task in achieving this aim of automated scaling is to forecast future workload, using the data on past workload history, so that enough VMs may be setup in advance to match the anticipated change in workload trend. Auto-scalers which provision resources by predicting the workload are categorized as proactive auto-scalers. This work aims at proposing a Machine Learning (ML) based workload prediction model for proactive auto-scaling of cloud resources. To this end, we demonstrate the use of Support Vector Machine (SVM) for Multi-class classification to predict the future workload by using a real cloud dataset. The results found are plausible as they indicate efficient and accurate workload prediction.