| With the surging amount of data produced by the whole society and the prevailing pay-as-you-go mode in cloud service,the size of data center is enlarging rapidly nowadays.The simultaneous operation of a targe number of physical nodes in server cluster brings significant power consumption to data center-At present,high energy consumption not only becomes a heavy financial burden for operators of data center,but also hinders the performance of hardware and software system.In order to realize green computing,energy-saving and emission-reduction,healthy development of cloud computing,the energy consumption issue in data center becomes increasingly prominent and urgent to be solved-Focusing on the energy consumption optimization issue from the SDCloud view,this thesis not only makes a thorough research on energy consumption in the processes of VM Consolidation and VM Migration but also investigates relevant problems such as resource contention,performance interference and workload fluctuation,etc.In this thesis,the energy and performance trade-off is the prime optimization objective in the design of specific strategies and algorithms.This thesis always pays attention to the possible deviations of theoretical models in practical application scenarios.All of these findings help achieving more accurate and more flexible energy conservation solutions due to the improved theoretical model and abundant experimental verifications.Specifically,the research achievements and innovations in the thesis are listed as follows:(1)In this thesis,we not only discuss the details of architecture and modules in SDCloud but also analysis the underlying challenges of the energy consumption issues in cloud data center.All of them are relevant to the design and implementation of consequent strategies and algorithms in this thesis.(2)A multi-dimensional resource supply-demand model is proposed in this thesis.In this model,both thc scarcity of resource in a cluster and the dynamic adjustment of-supply-demand of resources are taken into full consideration.By leveraging this model,this thesis also presents a new strategy with proportional constraints among different resource dimensions.This strategy can coordinate the consumption ratio of each resource dimension in the VM Consolidation process which helps to avoid performance bottleneck due to excessive consumption of resource.(3)The problem of performance degradation during the high resource utilization is raised whiclh is also the shortcoming of above mentioned strategy.From the perspectives of measurement model,coping strategy and monitoring method,relevant discussion and consequent research are conducted on the resource contention and performance interference issues which are prevailing in physical server and virtual machine.(4)Targeting at the side effects on the VM Migration from the fluctuation of dynamic workload,this thesis investigates the regularity of the workload fluctuation in cluster,taking the open source monitoring data from Google as the experimental dataset.Based on the traditional methods of time series analysis,a new comprehensive scheme is proposed which includes data processing and workload forecasting.In this scheme,the workload fluctuation can be perceived effectively thus corresponding resource threshold can be adjusted dynamically.Then abundant experiments are carried out which show that the scheme is efTcctive to deal with the energy and performance trade-off problem,(5)For the current approach in which single threshold is applied in VM Migration,this thesis investigates and discusses on the phenomena of over-migration and delayed migration.By introducing knowledge on rough sets,the two-class decision-making problem with single threshold is improved into three-class virtual machine migration decision problem.In addition,we also incorporate the effects of VM migration into the final cost function.With the overall cost of migration and Bayesian decision theory,the relevant decisional rules for VM migration can be established accordingly. |
PDF Full Text Request