Consolidation of Virtual Machines Using Stochastic Local Search

Abstract

A modern cloud data center is represented as a complex system where virtual machine consolidation and scheduling influence directly the cloud cost and performance. The virtual machine consolidation is the subject to many constraints originating from multiple domains, such as the resource requirements, user Service Level Agreement (SLA) compliance, security requirements, availability requirements, and other. Properly defined resource management methods and algorithms allow to achieve execution efficiency, SLA compliance, utilization of resources, energy saving, and the increasing profit of cloud providers. In this paper, the authors propose two versions of the Optimization using Simulated Annealing (OSA) algorithm to solve dynamic virtual machine consolidation problem. The virtual machine consolidation problem is considered as a multi-dimensional vector bin-packing problem. The authors take into account that the properties of items can be changed, new items may be requested to be deploy, and existing items may need to be reassigned to bins. Other constraints should be taken into consideration to solve virtual machine consolidation problem such as balanced load of resources of each physical machine, the limitation on maximum number of simultaneous migrations per physical machine, hardware constraints and other. The configuration of the system, the function for obtaining new configuration, the objective function for the optimization problem are determined for the proposed algorithms. The evaluation results show, that using OSA algorithms the simulated data center consumes almost the same amount of energy as while using a not optimized algorithm. On the other hand, the OSA algorithm with constraints allows to decrease overall performance degradation by virtual machines due to migrations, as a result, SLA violation is decreased. Furthermore, both OSA algorithms allow to reserve some resources of physical machine to react to increasing random resource demands in the nearest future.