Research On VM Management Mechanism In Cloud Data Centers For OpenStack

As cloud data centers become more and more mature,cloud computing has gradually become the dominant computing paradigm in the information technology industry,but this also brings the problem of high energy consumption.In the cloud data center,the resource usage of most physical machines(PM)is very low,causing high waste of energy.Virtual machine(VM)consolidation technology has proven to be an efficient way to save energy.This technology places VMs in the data center on as few PMs as possible,thereby switching PMs without VMs to sleep mode to save energy.However,current VM consolidation strategies are only outstanding for data centers running specific applications,such as CPU-intensive applications or network applications.Therefore,it is necessary to design a universal VM consolidation framework for a universal cloud data center to support PMs with different real-time characteristics.OpenStack NEAT is a mature VM consolidation framework integrated into OpenStack,which decomposes VM consolidation process into four sub-problems:(1)host underloaded detection;(2)host overloaded detection;(3)VM selection and(4)VM placement.However,this framework also has flaws.First of all,the framework cannot run on Otaca version or earlier versions of OpenStack,and the data used by the components in the framework may not be the latest.Secondly,the components of the framework cannot perform adaptive algorithm selection for PMs with different real-time characteristics,and the performance of the algorithms implemented by the framework is poor.Finally,the scalability of the framework is very poor,the addition of custom algorithms requires much review and modification of the framework source code,and the operation of the framework is complicated.In order to solve the problems above and make VM consolidation be adaptive to PMs with different real-time characteristics,this study designed and implemented an intelligent NEAT framework(I-NEAT).First,the I-NEAT framework fixes vulnerabilities in the OpenStack NEAT framework and designs algorithm libraries for each sub-problem with many new algorithms.In addition,based on the existing OpenStack NEAT framework,I-NEAT framework adds two additional components:intelligent scheduler and framework managers.The intelligent scheduler usingreinforcement learning is deployed on each compute host and is responsible for providing load detection algorithms for local manager according to the current real-time characteristics of the compute host,thereby implementing adaptive VM consolidation.The framework manager is deployed on controller host.It designs open templates for each sub-problem and provides upload operations for custom algorithms,which greatly improves the scalability of I-NEAT.Also,the framework manager provides service management and configuration management.Finally,this study conducted a verification experiment on a small-scale system to evaluate and analyze the performance of new algorithms and the intelligent scheduler using reinforcement learning.Performance metrics includes: AITF,AOTF,VM migrations,and energy saving.In order to make the experiment reproducible and convinced,this study uses workload,which is from cloud data center in real world,as simulated load,and fixes the mapping between load traces and VMs.Experimental results show that the performance of I-NEAT using new algorithm is significantly better than NEAT,and the intelligent scheduler can further reduce energy consumption and better avoid SLA violation.At the same time,he learning ability of the reinforcement learning can make the intelligent scheduler effectively learn strategies in a short time for efficient VM consolidation,to make the I-NEAT be adaptive to PMs with different real-time characteristics.