Research And Improvement Of Load Balancing Algorithm Based On Nginx High Concurrency Server

With the rapid development of the Internet and the huge amount of information it carries,the speed of information exchange is rapid and the degree of freedom is high,and the global information sharing has been realized.But at the same time,the increase of users leads to the surge of page views,which greatly increases the pressure on the network server and puts forward higher requirements on the server performance.Due to a large number of high concurrent access to the backend server pressure problem to be solved,thus arises at the historic moment of server cluster system and load balancing technology to a large extent alleviate the problem,and these technologies improve the performance of the server on one hand,on the other hand in a large amplitude reduced overhead needed to improve server performance.Now there are a lot of applications on the server load balance technology,some allocation strategy is insufficient,may affect a busy server processing tasks at the same time also received a large number of access request,and the part of the server is idle undistributed task to the request,this phenomenon will affect the entire cluster of servers for high response speed of concurrent access,which makes the user experience worse.To solve the problem of unreasonable allocation of load balancing strategy in server cluster,this paper studies the load balancing algorithm based on Nginx high-concurrency server,whose main work is as follows:(1)Time series prediction research based on server website visits.The main tasks are as follows: in view of the problem of server cluster performance degradation or even downtime caused by the surge of page views during the peak period of website visits,we can predict the past page views of the website and adjust the nodes of the server cluster according to the predicted results to improve the stability of the cluster.The network access data set provided by Kaggle competition "Web Traffic Time Series Forcasting" is used to analyze the data and determine the appropriate prediction model.The Time Series prediction algorithm is used to predict the request volume of network users,and the algorithm was improved and analyzed by simulation experiment to improve the prediction accuracy.(2)Forecast the overall trend of residual performance changes of each node in the server cluster in a day.Select appropriate indicators as the basis of collecting node load information,and in each node load information continuously collect data.The residual performance data of each node is taken as sample data,and the prediction model of the residual performance of each node is established.Aiming at the problem that the error of multi-step prediction of this model increases gradually,the historical time series data is optimized to change the result of multi-step prediction of the model to that of single-step prediction,so as to reduce the prediction error and better predict the change trend in a day.The prediction model was improved,the algorithm before and after the improvement was compared in simulation experiments to verify the effectiveness of the algorithm,and the prediction results were used as the basis for weight adjustment in the subsequent load balancing strategy.(3)Load balancing research based on Nginx high concurrent access.To solve the problem that the existing weighted polling algorithm of Nginx server is unreasonable,an improved dynamic load balancing algorithm is proposed to allocate the user's request tasks.The residual performance prediction results of the nodes are used to determine whether the current load balancing scheme should be adjusted.By referring to the principle of TCP congestion control,the weight of each server node is adjusted to achieve the purpose of controlling the load state of the nodes,and the Linux server cluster system is set up for the experimental analysis of load balancing algorithm.