Research And Improvement Of High-Concurrent Web Server Load Balancing Strategy Based On Nginx

In the process of large-scale web concurrent access,multi-server cluster technology is generally adopted to ensure the timely and accurate response of user's requests.As the core of its work,load balancing algorithm is responsible for allocating a large number of concurrent requests from clients to back-end servers,whose performance directly affects the working effect of server cluster.In the current implementation of concurrent Web server,Nginx has a strong ability of concurrent processing and good scalability,so it has been widely used.In this paper,based on the research of Nginx load balancing algorithm,two improved algorithms are put forward,and the specific work contents are as follows.First,based on the analysis of Nginx source code,this paper studied the Nginx built-in load balancing strategy and found its strategy is relatively unitary,The user requests were allocated dynamically can not based on the real-time load information of the backend server nodes.This paper presents a dynamic load Balancing algorithm(HDWP)based on RoundRobin,which integrates the characteristics of multiple service environments and the necessity of heterogeneous clusters.This algorithm collects and processes the load request from the client in real time,the request is classified,and converted into the corresponding weights,then the user's request is allocated according to the size of the weight value.tested the algorithm and Nginx built-in three load balancing strategy through established heterogeneous clusters of in the Linux system,the experimental results show that the algorithm is better than the Nginx built-in load balancing strategy in the average response time and the actual concurrent number of two performance indices.Second,based on the HDWP algorithm,combined with the theory of Markov chain and queueing theory,this paper proposed the HDWPM algorithm of a queueing model based on multidimensional Markov prediction.The algorithm mainly made Markov predictions of the load request from the client.Judging it whether caused every business clusters overload,if it leads to the business clusters overload,then not response to the request.Otherwise,the request is scheduled within each business cluster based on the queue length before each server and the load of the back-end server;The test results show that the algorithm shortens theaverage response time to some extent and improves the actual concurrent connection number of the user.