Handling hierarchy in cloud data centers: A Hyper-Heuristic approach for resource contention and energy-aware Virtual Machine management
Jiayin Zhang,Huiqun Yu,Guisheng Fan,Zengpeng Li,Jin Xu,Jun Li
DOI: https://doi.org/10.1016/j.eswa.2024.123528
IF: 8.5
2024-09-01
Expert Systems with Applications
Abstract:For cloud data centers, a performant yet energy-efficient operation is critical for service quality and experience. The growing demand for cloud-based services has led to the development of large-scale hierarchical data center structures, characterized by horizontal expansion and vertical hierarchy, leading to challenges in managing Virtual Machines (VM) at a granular level. The hierarchical arrangement can increase the risk of deployment failures, often stemming from inadequate computational resources on physical hosts, even when the cluster-level resources seem sufficient. While substantial work has gone into managing VMs at the physical host level, there remains a dearth of research under hierarchical data center configurations. To fill the research gap, we address the hierarchy in cloud data centers with a novel two-stage approach named VMM-HHGT, aiming at suppressing VM deployment failures, while balancing the energy consumption and computation resource contention. VMM-HHGT comprises a Hyper-Heuristic-assisted broker (VMM-HH), which can learn the workload patterns and hardware configurations to generate cluster-selection heuristics. An offline training process is incorporated for continuous heuristic evolution with zero overhead on decision-making. Besides, a Game-Theory-assisted hypervisor (GT) is designed for inter-host live VM migration for fine-grained balancing of energy consumption and resource contention. Extensive experiments with traces from real-world VMware data centers show that VMM-HHGT achieves a higher deployment success rate compared to the state-of-the-art approaches, with a well-situated performance in energy consumption and resource contention.
computer science, artificial intelligence,engineering, electrical & electronic,operations research & management science