Research On Key Technologies Of Application-level Load Distribution System

With the development of the Internet era,a large amount of network data interacts on the Internet.In the process of receiving,processing,and forwarding network traffic,each network node may have current routing problems,data request address problems,data keyword problems,etc.This causes the traffic to be out of proportion in the process of forwarding and scheduling,that is,the traffic is sent to some devices more,which causes other devices to be relatively idle.This phenomenon is called traffic skew.Load balancing technology is a common method to solve this traffic tilt phenomenon.With the deepening of load balancing technology research and the combination of technologies in different fields,industry scholars have proposed many efficient and stable models,which make the load balancing technology mature..The content to be discussed in this paper is the load balancing technology based on application layer data feature analysis.Firstly,the optimization problem of load balancing model in traffic scheduling and the real-time problem of load node status are addressed.A load balancing method based on DPI prediction is proposed.By analyzing the application layer protocol samples,the application layer protocol characteristics are determined and the regression equation is established.Then the traffic size is estimated.A dynamic load balancing model based on prediction is constructed by combining the predicted value with the node state of dynamic detection.The model can predict the status of load nodes before the balanced traffic dispatch,so as to optimize the load situation of nodes.In the experimental part,by comparing with the classical load balancing algorithm,it is verified that the load balancing model based on DPI prediction has stable and efficient load balancing effect.Secondly,in a larger network,when using load balancing device cluster to distribute traffic,there exists the problem of data skew caused by unbalanced traffic among load balancers.Based on the path calculation problem in graph theory,a load migration model based on dynamic detection is proposed in this paper.The model uses Floyd-based dynamic path generation algorithm to construct the path.Combining with the application principle of Hello message in OSPF,the model transmits and updates load balancing information.When data skew occurs,traffic can be migrated to the peripheral idle devices to achieve the goal of load balancing.In the experimental part,a distributed Map Reduce model is constructed to simulate the cluster environment,and the data flow with specified key skew is used to carry out the experiment.By comparing the load effect of the connection matrix method and the algorithm proposed in this paper,it is proved that the method can effectively reduce the data skew phenomenon of load nodes.