Research On Dynamic Allocation Technology Based On Nginx Load Balancing

At present,with the high popularity of Internet technology,explosive web concurrent access requests have brought huge challenges to network bandwidth and server performance.It is difficult for a single server to respond to user requests quickly.Although cluster servers can provide uninterrupted services with higher performance than a single server,they can not allocate requests reasonably and effectively and improve the overall performance.the resource utilization of each system,so load balancing technology emerges as the times require.As a representative of excellent load balancing,Nginx has the advantages of low resource consumption,high concurrency and easy expansion.Therefore,it is necessary to study the dynamic allocation technology based on Nginx load balancing.Firstly,the author analyzed the modular structure of Nginx's source code,the processing flow of Nginx's HTTP requests and its own algorithm,and found that its own algorithm has some shortcomings,such as inability to feedback the server's performance in real time,dynamic modification of the server's weight and incomplete collection of server indicators.Then,according to the above shortcomings,an innovative weighted polling algorithm is proposed.Dynamic load balancing algorithm.In addition to adding load information real-time collection module,load information processing module,improved entropy method to determine load weight coefficients,and experimental determination of upstream server weight modification threshold and server load information collection cycle,the algorithm also improves the Nginx kernel and configuration file.Finally,in the experiment,the response time,throughput rate and maximum concurrency are selected as the evaluation indexes of the algorithm to comprehensively test and analyze the improved algorithm.In this thesis,we use seige performance tools to evaluate the performance of weighted polling,IP-hash,third-party fair and the improved algorithm proposed in this paper.Experiments show that when the concurrency of dyn reaches its peak and saturation state,the response time of dyn algorithm is 2.39% and 11.9% shorter than that of the other three algorithms,and the maximum concurrency is the best.IP-hash improved by 3.90% and 16.9%.Throughput of IP-hash improved by 0.52% and 7.77% compared with weighted polling RR of the optimal algorithm.Therefore,the improved algorithm in this thesis can achieve better load balancing effect than other algorithms in various performances.