Design And Implementation Of Streaming Media Real-time Server Cluster

With the continuous development of network applications,server load pressure increases rapidly.Under this background,this paper takes the real-time server of streaming media as an example,deeply studies the server load technology,designs and implements a real-time server cluster of streaming media.To ensure that it can respond to a large number of client requests within a specified time,and when a small number of servers fail,the rest of the servers can still ensure real-time,each server's hardware resources can play a maximum role.Usually there are three major problems in server cluster under load pressure: multi-parameter load balancing of server resources,bottleneck of cluster export,and restart maintenance of server after running for a period of time.In this paper,the three problems are studied in depth,and the corresponding solutions are put forward.The contents are as follows:(1)For the load balancing problem,according to the method of comprehensive evaluation,the multi-resource factors that affect the performance of servers are evaluated as a whole.Firstly,a set of elements and a set of judgements are set up for each server in the cluster.Then,the resource weight vector is calculated by the method of experimental evaluation.Thirdly,according to the basic principle of fuzzy comprehensive evaluation,the mapping matrix from resource weight vector to service capability vector is obtained.Finally,the best performance servers are selected from the server cluster as the result of this request redirection.(2)To solve the bottleneck problem of cluster export,a compensation repair model of server cluster based on fractal geometry is proposed.Firstly,it makes the network topology of cluster have local and global similarity.Secondly,any server in the cluster can get all available server status in the cluster.Thirdly,the client can switch to any server.Decentralization of this model can eliminate the bottleneck of equalizer.Compensation capability can compensate fault server dynamically.Regenerative compensation capabilities can dynamically add servers or even sub-clusters.(3)To solve the problem that the server must restart after running for a period of time,a fast search algorithm in network programming is proposed.Firstly,we use the network programming interface inherited from UNIX and the completion queue interface to achieve efficient thread pool.Secondly,the operation objects that need to be accessed when the asynchronous network operation is completed are inserted into the completion queue.Thirdly,when the network operation is really completed,it finds the operation object quickly and finds the operation object to complete the processing.This model improves the efficiency of IO access,avoids a large number of memory fragments,reduces the CPU load of thread switching scheduling,and reduces the cost of communication object search.It reduces the server load and prolongs the fault-free running time of the server.The experimental results show that the proposed method solves the multi-parameter load balancing problem,the bottleneck problem of cluster export and the restart maintenance problem of the general method,and can improve the real-time performance of the server cluster.