ENERGY EFFICIENT HEURISTIC FRAMEWORK FOR VIRTUAL MACHINE PLACEMENT IN CLOUD DATA CENTERS

Abstract

Cloud data centers are growing rapidly in both number and capacity to meet the increasing demands for highly-responsive computing and massive storage. Cloud is a virtual infrastructure that is accessed or delivered with a local network or accessing the remote location through internet. As a cloud is realized on large-scale usually distributed data-centers, it consumes an enormous amount of energy. Several researches have been conducted on Virtual Machine (VM) consolidation is an emerging solution for energy saving. Among the proposed VM consolidations, Open Stack Neat is notable for its practicality. OpenStack Neat is an open-source VM consolidation framework that can seamlessly integrate to OpenStack, it can be configured to use custom VM consolidation algorithms and transparently integrates with existing OpenStack deployment without the necessity of modifying their configuration. The framework has components for deciding when to migrate VMs and selecting suitable hosts for VM placement. It focuses on minimizing the number of servers. However, the solution is not only less energy efficient but also increases Service Level Agreement (SLA) violation and consequently cause more VM migrations. Therefore, in this research work we proposed energy efficient heuristic framework for VM placement to address the problem of allocation and consolidation of Virtual Machines by modifying the bin-packing heuristics with the power-efficiency parameter. In addition to that, we introduced two solutions: First, in the overloaded host decision step, the algorithm check whether a host is overloaded with SLA violation or not based on the overload threshold and specification of the active hosts. Second, in the underloaded VM migration step, this study puts forward a minimum power policy then power off the target host. Finally, to evaluate the proposed framework we have conducted experiments using CloudSim on three cloud data-center scenarios: default, heterogeneous and homogeneous. The workload that run in the data-center scenarios are generated from traces of PlanetLab and Bitbrains clouds. The experimental evaluation shows that our framework minimizes the energy consumption by 62.3% and reduce SLA violation and number of VM migrations by 75.73% and 68.73% respectively compared to the existing framework.