Proactive Workload Prediction and Resource Management in Hybrid Cloud using Machine Learning Techniques

Abstract

Cloud Computing (CC) paradigm has improved information and communication in recent years and provided a backbone to modern infrastructure. CC enhances the services of organizations such as Government, industries, and academia with a payas-you-go model. More than 60% application workload is migrated to CC. The applicationshosted on CC heavily use resources and generate more traffic, specificallyduring specific events. The management of resources is one of the issues in CC. To achieve better quality in service provisioning and avoid Service Level Agreement (SLA) violation, the elasticity of resources is a major requirement in CC. The hybrid cloud model excels in resource requirements with private and public cloud services to deploy elasticity applications. The resource monitoring and prediction improve the resource management policy with elasticity. For elasticity, a traditional adaptive policy implements threshold-based auto-scaling approaches that are adaptive and simple to follow. However, such a static threshold policy may not be effective during a high-dynamic and unpredictable workload. An efficient auto-scaling technique that predicts the system load is essential. Balancing the dynamism of load through the best auto-scale policy is still a challenging issue. This research work addresses resource prediction mechanisms to handle workload demands in CC through ML techniques. This work explores how these techniques can be adapted to resource management problems to increase resource availability and reduce SLA violations of Cloud data centers while simultaneously satisfying application QoS requirements. The data center parameters such as CPU utilization and users’ requests are analyzed and suggest an algorithm using Machine learning and Queuing theory concepts that pro-actively indicate an appropriate number of future computing resources for short-term resource demand. The experiment shows that the suggested model enhances the elasticity of resources with performance metrics. The suggested approach is evaluated against other baseline approaches. Overall, we conclude that a machine learning-based auto-scale approach to optimize resource prediction aids a Hybrid Cloud resource management system with fewer SLA violations.