Research On VM Placement Optimization Algorithm Based On Machine Learning

Currently,cloud computing technology has been developing rapidly,and virtualization technology,as the key of cloud computing,has been one of the hot issues of research.There are multiple aspects in virtualization technology such as network virtualization,virtual machine placement,migration of virtual machine,etc.In the cloud data center,it is common to place multiple virtual machines on the same physical server to improve the resource utilization.Reasonable and efficient virtual machine placement algorithm can improve the operation efficiency of data center and save operation and maintenance cost.Therefore,the placement of virtual machine needs to consider many factors,including reliability,energy consumption,network resource consumption,etc.There have been several studies based on traditional or heuristic algorithm in the last few years.Some researchers have already applied machine learning techniques to virtual machine placement problems,but most of them only regard machine learning as an auxiliary strategy,such as using machine learning techniques for demand forecasting,and few of them apply machine learning techniques to decision making directly.In this thesis,based on previous works,new algorithms with machine learning technology have been proposed to process virtual machine placement problem.The main work of this thesis includes the following aspects.(1)Virtual machine placement algorithm based on clustering.Clustering algorithm is common in machine learning technology.And it gathers data to form clusters with a certain rule.The problem of virtual machine placement can be seen as a problem of clustering a series of virtual machines.A virtual machine placement algorithm based on clustering has been proposed to solve the virtual machine placement problem under resource constraints.The constraints in this problem include resource demand such as CPU and memory,and communication demand between virtual machines.Based on clustering,this algorithm will select the virtual machine with the most traffic flow as the core of clustering,and design reasonable clustering rules to build this virtual machine placement algorithm.(2)Virtual machine placement algorithm based on Deep Q Network(DQN).The success of AlphaGo makes DQN,the technology behind of AlphaGo,reciving widespread attention and developing rapidly.DQN is one type of deep reinforcement learning algorithm.Deep reinforcement learning has powerful self-learning ability and can solve decision-making problems under complex environment.In this article a virtual machine placement algorithm based on DQN has been proposed,which can self-learn without prior knowledge and the trained model can process the problem of virtual machine placement.And then there is an improvement scheme based on multi-model decision-making which using multiple models to make decisions at the same time,and choosing the best one from all of the decisions as the final decision.Finally,the simulation results show that our algorithms are excellent.