【作者】 孙翔；

【导师】 袁凌；

【作者基本信息】 华中科技大学 ， 计算机软件与理论， 2017， 硕士

【摘要】 虚拟机整合为IaaS云数据中心中提升能源效率和服务质量提供了一种有效途径。大部分研究者将虚拟机整合阐述为装箱问题,装箱问题为NP-Hard问题。由于没能将虚拟机负载视为时间序列进行分析,当前虚拟机整合的效果不佳。一方面,虚拟机短期负载往往具有明显的趋势性,通过时间回归法可以定量分析虚拟机短期负载的变化趋势,结合负载变化趋势和当前负载能够有效提高从过载物理机选择待迁移虚拟机的准确性。基于此思想,设计了基于负载增量预测的迁移虚拟机选择策略(Loading Increment Prediction,LIP)。另一方面,如果将负载序列具有互补效应的虚拟机整合到同一个物理机,将会有效提高物理机负载的平稳性,从而减少由于物理机发生资源竞争而引起的服务性能下降。基于此思想,设计了基于负载序列预测的虚拟机迁移点选择策略(Saturation Increase Rate,SIR)。而所设计的基于负载预测的虚拟机整合算法即为LIP策略和SIR策略的有机结合。最后,通过使用真实的负载数据在云计算模拟仿真器CloudSim中进行仿真实验来评估所提出的策略和算法。LIP相比最小迁移时间策略在能源消耗、SLA违约率和虚拟机迁移代价指标上分别减少了近27%、46%和42%。SIR相比最小能耗增加策略分别减少了近25%、32%和65%。从而说明所设计的基于负载预测的虚拟机整合算法能够有效提升数据中心的整体服务性能和能源效率。更多还原

【Abstract】 Virtual machine consolidation(VMC)provides an effective way to improve energy efficiency and quality of service in IaaS Cloud Datacenter.Most researchers describe VMC as a packing problem.As a well-known NP-Hard problem,which usually use heuristic algorithms to solve.Failing to treat the load of virtual machine(VM)as a time series for analysis,the efficiency of current VMC is not good enough.On the one hand,the short-term load of VM often has obvious trend,time regression method can be used to analyze the increase trend of VM load and quantities the load increment.Then,this paper proposes a migration VM selection strategy named Load increment Prediction(LIP)based on load increment prediction.LIP strategy makes it more accurate to select the VM from the overloaded physical machine(PM)by combining the current load and load increment.On the other hand,if the VMs whose load series complementary each other are consolidated into the same PM,the smoothness of the PM load will be improved.Therefore,it is possible to reduce the SLA violation and VM Migration due to the resource competition of the PM.Based on this idea,this paper proposes a VM migration point selection strategy named Saturation Increase Rate(SIR)based on load sequence prediction.The VMC algorithm based on load prediction proposed in this paper is the organic combination of LIP algorithm and SIR algorithm.Finally,we evaluate our strategies and algorithm by simulating experiments in cloud computing simulator CloudSim using real load data.The experimental results show that LIP algorithm and SIR algorithm can effectively reduce the cost of VM migration and SLA violation due to the resources competition of PM.更多还原