Research And Application Of High-Concurrent Access Server Based On Nginx

With the development of the Internet,the maturity of the mobile Internet,the everincreasing number of users,and the diverse needs of users,the traditional server architecture has been unable to withstand concurrent requests from massive users.This article studies the Nginx server,masters the system architecture of Nginx,and analyzes the reasons why Nginx can handle massive concurrent concurrency.Through analyzing the demand of the shopping spree scene,this paper summarizes the objective reasons for high concurrent access,analyzes the factors that affect the server's ability to concurrently,and proposes strategies and algorithms to improve the server's ability to perform concurrently.The demand for shopping spree scene is very extensive in China,whether it is a train ticket buying or grabbing a red envelope is its application scenario.There are many reasons for the high concurrency,and the means to reduce the amount of concurrency are also varied.There are many factors affecting the concurrent volume increase.There are design issues,hardware and software performance issues,and algorithm performance issues.They can be improved.This paper presents a load balancing algorithm based on predictive model and independent training nodes by studying the existing load balancing technology and machine learning technology.The algorithm uses machine learning technology to obtain a response time prediction model by training historical data to predict the response time of new requests.According to the estimated response time of each server node,the requests are distributed to server nodes with the least response time.Improve the balance of request allocation in the cluster and improve the efficiency of the cluster.In order to reduce the performance pressure of the machine learning algorithm on the load balancer,this paper uses independent nodes to train the prediction model,and reduces the amount of computation in the machine learning process by reducing the scale of characteristic parameters.This paper reduces the network traffic within the cluster from the algorithm level as much as possible to further reduce the response time.This paper compares three kinds of machine learning algorithms,and shows that the algorithm can reduce the average response time of the system in high concurrency scenarios of small and medium clusters.