An Anti-overload Model For Openstack Based On Effective Dynamic Migration.

As an emerging technology, cloud computing has revolutionized information technology industry and continues to draw a significant attention from both public and private sectors. As rapid application of this evolving technology increases, maintaining performance with respect to factors like Quality of Service (QoS) and Service Level Agreement (SLA) is vital. These factors need to be considered when we choose among cloud computing service providers. Virtual Machine(VM) migration is one of the main operations performed to reduce aggressive consolidation which otherwise may lead to performance degradation of applications.In this thesis, we propose a dynamic approach for live migration to support an Infrastructure-as-a-Service (IaaS) and avoid physical machine overload. This approach significantly contributes to improvement of the rapidly growing and the dominating cloud computing open source platform known as OpenStack. Careful studies on OpenStack have enabled us to observe an aggressive overload ratio exhibited by the core project of OpenStack. This problem can be reduced by conducting cost aware dynamic live migration of virtual machines as required and at an appropriate time to reduce violation of QoS and SLA.This work employs a single exponential smoothing model to estimate the physical machine load at the next interval. Furthermore, we proposed an efficient distribution algorithms based on adaptive threshold policy which is preferred for unstable workload environment. The algorithms which are distributed on local hosts are efficiently used to detect the overload before its likely occurrence. Then the fitting VMs are chosen to keep the load under threshold policy. Moreover, a different algorithm at the controller effectively allocates the VMs to the lightest target Physical Machine (PM). We believe that this work contributes to the improvement of OpenStack cloud platform in terms of managing and redistributing resource which fits extensible sets of management objectives.The implementation consists of distributed algorithms, a load generator and monitoring. Experimental Results indicate that our system succeeded in avoiding the overload at suitable time, and the simulation results demonstrate that our approach achieved efficient reduction in SLA violations and the number of VM migrations.