Research On Data Center Network Design Based On Relieving-Pooling Algorithm

The continuous development of digital economy continues to push up the demand for data centers of enterprises.At present,the mainstream data center network construction modes in the industry are self-building and operating own computing capabilities,self-building but operating by third parties which is colocation and purchasing capabilities on could.These three models are different in cost composition and service quality.how to optimize data center location,resource allocation and demand allocation according to the characteristics of each model is a difficult problem for enterprises.In view of these three modes,based on the queuing theory and the classical model of service facility location,this paper establishes the data center network design optimization model respectively,develops the algorithm to solve,and carries on the cross-pattern comparison.Firstly,the self-building and self-operating mode is studied.By analyzing the direct cost of data center network construction and the indirect cost related to service delay and data security in operation,and considering various decisions such as data center location,resource allocation and demand allocation,a mixed integer programming model is established to minimize the comprehensive cost.By analyzing the characteristics of the model,in the framework of Lagrangian relaxation algorithm,a relieving-pooling algorithm based on greedy rules is designed and developed to solve the model effectively and efficiently.The numerical experimental results show that this algorithm can obtain better quality solution than Lagrangian heuristic algorithm and genetic algorithm to solve similar problems in the literature.Then,according to the respective characteristics of entrusted colocation mode and purchasing service mode,the corresponding mixed integer programming model is extended by using similar modeling method,and the solution is solved by using the developed relieving-pooling algorithm based on greedy rules.the optimal design of cross-schema data center is compared,and the effects of key parameters such as demand arrival rate and demand resources on pattern selection are discussed.Finally,the research results are applied to the specific case of C enterprise data center network design,which shows that the proposed model and algorithm can effectively solve the practical problems of the enterprise.