Research On Energy Conservation Of Geo-Distributed Data Center

In recent years,the rapid development of information technology,data center has been gradually development and progress.The cloud-based distributed data center came into being,the most important of which is the geographically distributed data center.The many advantages of geo-distribution can provide users with flexible services.However,the resulting energy consumption problem is also increasingly serious.Related theoretical research is mainly focused on energy-saving optimization of data center servers,and a small amount of research is optimized for energy-saving of data center network transmission.The problem model of combining the above two optimizations is seldom exp lored.In a geographically distributed data center,the core goal of energy cost optimization in geographically distributed data centers is to reduce the electricity prices cost.While optimizing the cost of energy consumption,it will restrict each other with many other constraint targets.Based on the above considerations,this paper establishes a model of energy consumption cost optimization with the cost of electricity price as the objective function.Then the Lyapunov optimization of the problem model is carried out,and the carbon emission constraints which are difficult to calculate directly are transformed into the stability of the virtual queue.Then the transformed model is solved by using the generalized Texaco decomposition algorithm.On this basis,the factors of heterogeneous load and heterogeneous data center are added into the energy cost optimization model which can accommodate heterogeneous factors.Finally,a numerical simulation of the proposed algorithm is carried out in MATLAB,and it is shown that the algorithm proposed in this paper is more effective in saving the cost of energy consumption under the same experimental data.The mathematical modeling solution can solve the optimization problem,but it still exists the problem that there is no optimal solution in some cases and some cases can not establish the appropriate mathematical model.Using reinforcement learning,we can break through the limitation of problem modeling to a certain extent.We only need to give reasonable action,state,reward function design,transform the problem into Markov process,and put forward the corresponding algorithm for the problem.The current energy-saving research based on intensive learning does not take into account the fault tolerance of the system.The optimization algorithm based on reinforcement learning can deal with the change of system environment better.This approach which consider the processing of load redistribution can respond to the fault of geo-distributed data center nodes in time after fault.In this paper,an energy-saving load distribution strategy based on reinforcement learning is proposed.The energy-saving optimization and fault tolerance are considered comprehensively.The experimental numerical simulation is carried out to illu strate the effectiveness of this method in energy-saving effect and its superiority in fault-tolerant effect.