Research On Technology Of CDN Load Balancing

With the continuous development of Internet technology and the change of people's thinking mode,the network has become an integral part of the lives of nearly half of the world's population.The ever-increasing number of Internet users and the continuous extension of the Internet time have brought unprecedented traffic to the Internet and the communications industry.With the traffic surging at the same time,another problem comes with it,which is how to quickly respond to user access request.Content Delivery Network(CDN)is used to solve the contradiction that the network bandwidth and network access speed can not be proportional.During the process of CDN running,server load balancing is critical.Because each edge node has different service ability and different user's distribution,if there isn't a good scheduling mechanism,it can also cause accessing congestion of the edge node,and affect the user experience.Therefore,this paper will start from the practical application scenarios,and divide CDN into three parts of node selection,content distribution and data storage,and design the improved algorithm aiming at the existing problems and shortcomings of the load balancing technology,and compare with the existing load balancing algorithm.The main contents of this paper are as follows:In the aspect of node selection,the Load Balancing K-means algorithm is proposed for the global load balancing node selection.This algorithm can specify the quantitative proportion of the inner elements of clusters before clustering.Finally,we compare it with K-means to prove that this algorithm can realize clustering effect with load balancing.In order to solve the problem of local load balancing node selection,the Chameleon algorithm is improved.Specifically,in the Chameleon algorithm decomposition stage,the improved Stoer Wagner global minimum cut algorithm based on the minimum cut hypothesis is used.In the Chameleon algorithm merging stage,the outlier optimization algorithm based on kNN algorithm is used.Finally,the two improved algorithms are experimentally tested to prove the feasibility.In the aspect of content distribution,one responsive load balancing distribution algorithm is proposed to solve the problem that load is not balanced enough when using traditional load balancing distribution algorithm.Finally,the polling algorithm,the minimum response time algorithm and the responsive load balancing distribution algorithm are experimentally tested.The experimental results show that this algorithm has better load balancing effect on content distribution.In the aspect of data storage,one virtual node with equal division Hash algorithm is proposed to solve the problem of load imbalance in the data storage node additions and deletions scene.Finally,this algorithm is compared with the virtual node consistency Hash algorithm,and the result shows that this algorithm has a better load balancing effect in the data storage node additions and deletions scene.