Research And Application Of Load Balancing Algorithm For Streaming Media Server

With the continuous growth of streaming media business,the existing video service is difficult to meet the needs of users.Cluster video distribution technology and HLS dynamic adaptive streaming technology are effective methods to improve video service quality.Based on the study of domestic and foreign cluster load balancing algorithms,this paper proposes two improved cluster load balancing algorithms and designs and implements a multi-node HLS streaming media distribution system.The main work is as follows:(1)Introduce four main dynamic adaptive Streaming technologies: Microsoft's Smooth Streaming technology,Apple's HLS technology,Adobe's HDS technology,MPEG's DASH technology,Web server cluster technology,software and hardware load balancing technology and dynamic and static load balancing algorithm.(2)Custer load balancing algorithm(RSC)based on residual service capacity and cluster load balancing algorithm(LBR)based on load balancing ratio are studied.Among them,the RSC algorithm USES the remaining service capacity of each node for task allocation and fails to make the load of each node balanced.LBR algorithm mainly considers the load balance of each node,and carries out task allocation according to the principle of equal load ratio of each node,without making full use of the remaining resources of nodes in the cluster.Aiming at the existing problem of the RSC and LBR algorithm,this paper puts forward a kind of based on the weighted residual service capacity and load balance than the cluster load balancing algorithm(RLS),the algorithm using stress testing method to evaluate node service capacity and load,introducing the dynamic feedback mechanism,made full use of the remaining service capacity of each node and the node load balance.Finally,three algorithms are simulated and the validity of RLS algorithm is verified.(3)The cluster load balancing algorithm(PKL)based on load weights is studied.PKL algorithm is based on the residual load distribution of each node without considering the difference of processing capacity of each node.Aiming at the existing problems of PKL algorithm,a cluster load balancing algorithm(DFL)based on node processing capacity and load combination allocation is proposed.In the algorithm,node processing capacity is measured by four indexes of CPU frequency,memory capacity,disk I/O rate and network bandwidth by ahp.Similarly,the load of nodes is also measured by the utilization ratio of CPU,memory,disk I/O and network bandwidth by the analytic hierarchy process.By introducing dynamic feedback mechanism,the nodes with normal current load are selected for task allocation according to their processing capacity.Finally,the two algorithms are simulated with opnet and the validity of DFL algorithm is verified.(4)An HLS video distribution system is designed and implemented,which is composed of a load balancer and three video service nodes.Under the coordination of the load balancer,three service nodes can coordinate video distribution,which improves the utilization rate of node resources and the service capacity of the system.The system has video recommendation,upload,query,classification,codec,slice and user information management functions.The system use LNMP framework to design and implement,and the service node use Nginx Web server to store video sharding file and video index file.The load balancer use the Nginx balancer to distribute video requests;MySql is used to store user management data,video file index information and load information.FFmpeg was used to encode and decode video,and slices were formed into TS fragment file and M3U8 index file.Finally,the implementation of DFL algorithm in the system is described.