Load-aware predictive auto-scaling framework for cloud environments

Abstract Cloud has become an increasingly popular computing paradigm because of its benefits in terms of scalability, flexibility and cost. To make cloud solutions competitive, it is important to exploit their elasticity and dynamically adjust cloud resources in a timely manner to cope with the incoming workloads. This paper proposes a load-aware predictive auto-scaling framework that tries to anticipate workload changes and ensure at the same time the desired utilization level of the cloud resources. To this aim, we forecast the future workloads and the cloud resources necessary to cope with these workload demands. The framework has been extensively tested in a simulation environment based on the CloudSim toolkit. Synthetic and real workloads characterized by different arrival patterns have been considered. The results showcase the effectiveness of the proposed policy in scaling cloud resources according to the predicted arrival patterns.