Learning-centric resource management in energy-efficient cloud computing

Abstract

107252-1.pdf Cloud computing has revolutionized the IT landscape by introducing a subscription-based model for computing resources. This on-demand resource delivery, coupled with a pay-as-you- go pricing structure, empowers users with unparalleled convenience and elasticity in resource scaling. This convenience has led to a rapid growth in cloud data centers, which are predicted to consume a significant portion of the world's electrical usage by 2025 [1]. This growth poses a substantial energy challenge, highlighting the need for energy-efficient cloud computing solutions [3]. Modern cloud data centers are complex, comprising heterogeneous servers, networking elements, power distribution systems, and cooling systems, all interacting in complex ways [16]. This complexity, coupled with the exponential growth of data centers and diverse workloads, necessitates research in energy-efficient cloud computing. Dynamic VM consolidation plays a significant role, Among various energy-saving techniques employed in cloud data centers, which includes migration of virtual machines (VMs) between servers to consolidate workloads and idle servers, thereby saving energy. However, existing dynamic selection algorithms for VM migration are heuristic-based and lack accurate utilization predictions, hindering their effectiveness. This research presents a novel approach to dynamic VM consolidation, leveraging a Machine Learning (ML) driven model. This model forecasts CPU utilization trends for individual VMs by analyzing their historical data. A VM selection algorithm is then proposed, utilizing the ML model to perform intelligent dynamic VM consolidation. The proposed algorithm's performance is assessed using a simulated cloud data center driven by real cloud workload traces, focusing on energy efficiency and customer Service Level Agreements (SLAs) violations, compared against existing heuristic-based algorithms.