Research On Intelligent Virtual Machine Management Based On OpenStack

As cloud computing is becoming more and more popular, it is more urgent than ever to improve the quality of cloud service, the utilization of datacenter resource and reduce the energy consumption, customer cost at the same time. It’s of great importance to do a series of further research into this subject and I present an intelligent virtual machine management model based on workload prediction to address the issues mentioned above.The workload of a cloud application may face sudden changes due to the complexity of cloud environment and the unpredictable behavior of users. This will lead to the drop of quality of cloud service and the satisfaction of customer. I come up with an efficient method to solve this problem by using the combination of virtual machine usage prediction and algorithms used in machine learning to find the bottle-neck element of virtual machine resources, such as CPU、RAM、 inbound bandwidth、 outbound bandwidth. After finding the bottle-neck resource, we can use an extensive of methods the scale the resource through virtualization interface. In this way we prevent the violation of SLO before it happen.This paper also used a virtual machine management model to take advantage of the regularity of long term workload changes. After predicting the request number of users, we can make suggestion on which virtual machines to use and which ones to return to cloud service provider to minimize the cost and preserve the resource demand. We also take server consolidation into consideration and come up with a multi-object optimization algorithm to minimize the use of physical server、 virtual machine migration number^ energy consumption and obtain a balance among these objectives. So we can use this algorithm to keep the use of physical machine、 virtual machine migration number and energy consumption at a relative relatively low level. This algorithm not only benefit the cloud service provider by lowering the cost of running datacenter, but also save some money for customers.The last part of this paper, we analysis OpenStack cloud platform and optimize the management process by using the combination of dynamic virtual machine configuration based on short term workload prediction and virtual machine management based on long term regular user request. This paper also shows the effectiveness by giving a prototype system based on JTangCMS.