Abstract
Cloud computing through the use of Virtual Machines (VMs) allows data center providers to modulate the computing resources allocated to users based on their needs. This assignment must satisfy user service level requirements (SLA) while reducing datacenter costs by maximizing resource usage and minimizing energy consumption. One powerful tool is the ability to redistribute the host load by reallocating the VMs that collapse the hosts. However, the migration process should be done cautiously, as the new VM allocation can negatively affect the VMs present in the target hosts, inciting unexpected new migrations. Furthermore, migrations are not free of charge as they require extra energy consumption and can also overload the interconnection network. This paper evaluates the effectiveness of combining Prophet forecasting and Bollinger Bands to predict cloud workloads, leveraging trend and seasonality analysis to improve virtual machine consolidation. To address this challenge, we propose a mechanism that prioritizes the migration of the most significant virtual machines according to their impact on host communications and energy consumption, together with a more realistic assessment of the impact of communications on overall power usage. The proposed approach was evaluated through an extensive analysis using cloud workloads from multiple providers (PlanetLab, Alibaba, Materna and Microsoft Azure) and compared against recent methods in the literature with similar optimization objectives. The results indicate that the proposed approach constitutes an effective and robust method for decision-making in VM migration processes across workloads in cloud computing environments.