Machine Learning Based Approach for Virtual Machine Migration

Abstract

Nowadays the utilization of Cloud computing in industry, government and academia are growing substantially, this is due to their ability to deliver resilient, robust, and scalable computational power. In cloud computing data centers, high-speed networks interconnect both virtual and physical computers. The dynamic provisioning of these systems is based on end-user computing resources requirements. However, high efficient resource utilization is yet far to reach, therefore the operational costs of these data centers are considerably high. Currently, systems allocate a maximum number of resources in a manner that attempts to ensure that all the Service Level Agreements (SLA) are satisfied. Virtualizations represents the core technology of cloud computing.  By creating multiple Virtual Machine (VMs) instances, it allows Cloud providers to manage its data center resources more efficiently and accordingly improving the utilization of resources. In addition to that, by using live migration the VMs can be dynamically consolidated on the minimal number of physical nodes according to their current resource requirements and maintaining SLAs. Therefore, non-optimized and inefficient VMs consolidation may cause performance degradation when an application encounters variable workloads. Thus, to ensure acceptable Quality of Service (QoS) and Service Level Agreements (SLA) a novel machine learning based technique for dynamic consolidation of VMs based on an adaptive prediction of utilization thresholds is proposed to satisfy the Service Level Agreements (SLA). Workload traces from Planet Lab servers have been utilized to validate the efficiency of the proposed technique with different of workload patterns.