Research On VM Consolidation Mechanism Of Cloud Data Center

Over the years,the advancement of cloud computing technology has accelerated the growth of cloud data centers both in scale and quantity.Yet,their further development and applications are hindered by high energy consumption and carbon dioxide emissions.VM consolidation mechanism based on virtualization technology is considered an effective solution to issues related to energy consumption.However,cloud data centers have a large volume of data and highly dynamic changes of load.Especially in the optimization of energy consumption,they should be meet different requirements for quality of service(QoS)by multiple applications simultaneously.In other words,they must not violate the service level agreement(SLA)signed between cloud providers and users.Therefore,how to minimize the energy consumption of cloud data centers under the premise of SLA has become a hot research issue in the academic and industrial worlds in recent years.VM consolidation mechanism means migrating VMs to fewer servers and switching idle servers to energy-saving mode to reduce energy consumption.Due to the high proportion of static energy consumption and the narrow dynamic power range of servers,VM consolidation mechanism can effectively improve resource utilization and reduce energy consumption in cloud data center s.Nonetheless,the excessive migration of VMs and reduction of active servers will lead to performance degradation of cloud data centers,thus negatively influencing user experience and operator income.To solve the above problems,this paper explores how to optimize VM consolidation mechanism with multiple types of loads by using the machine learning method in an Iaa S cloud environment under QoS constraints.The specific research work of this paper can be summarized as follows :First of all,aiming at VM consolidation of cloud data centers with Markov load,this paper proposes a VM consolidation algorithm based on a 3-order Markov chain model HS3MC(Host State 3-order Markov Chain,HS3MC).Using the 3-order Markov chain technology,the model predicts the state of hosts and puts forward dynamic threshold methods,thus effectively solving low performance caused by the mismatch between the current first-order Markov chain model and the load characteristics.Furthermore,both the VM selection algorithm focused on VM migration,CPU,RAM and the VM placement algorithm on the future state of the target host are developed.Afterward,a series of experiments are designed to verify the rationality of the model and to select the appropriate parameters for the algorithms.The experimental results confirm the role of the proposed model in effectively reducing the energy consumption and SLA violation level of the cloud data center with Markov load.Second,regarding VM consolidation of cloud data centers with general load,this paper proposes a VM consolidation algorithm based on a Bayesian classifier prediction model HSNBC(Host State Naive Bayesian Classifier,HSNBC)and a binary decision tree classifier prediction model HSBDTP(Host State Binary Decision Tree Prediction,HSBDTP).With the multi-point hybrid forecasting and decision method applied to the two models,effective methods of feature vector selection and label determination are designed and both the split rules and the VM placement algorithm on the future state of the target host are optimized.The experimental results show the great performance of the algorithms.On the whole,both the two models can effectively reduce the energy consumption and SLA violation level of cloud data centers.However,they have respective features.Specifically,the HSNBC model is better in performance,while the HSBDTP model is more simple and efficient.Thirdly,concerning VM consolidation of cloud data centers with high dynamic load,an online VM consolidation algorithm based on a linear regression model Robust SLR(Robust Simple Linear Regression,Robust SLR)is proposed.Instead of the traditional static resource reservation method,the model uses the dynamic resource reservation method to improve the overall performance through eight dynamic resource reservation strategies.The experimental results show that the proposed Robust SLR algorithm can effectively reduce the energy consumption and SLA violation level of cloud data centers with high dynamic load.Finally,aiming at the poor functionality and scalability of the existing VM consolidation platforms,this paper develops a dynamic VM consolidation platform based on Open Stack.As a transparent plug-in of Open Stack,the platform has complete functions and strong expansibility.It uses the deep Q-learning algorithm to realize intelligent state detection,allowing users to flexibly configure VM consolidation algorithms.Moreover,the platform can implement all VM consolidation algorithms developed in this paper.A two-month platform experiment is carried out with the installation and deployment of the platform on the blade servers,revealing the good performance of both the proposed platform and algorithms.