Research On Micro Service Load Balancing Algorithm Based On Improved Weighted Minimum Connections

With the rapid development of Internet technology,the scale of website applications continues to expand.Conventional software system architectures have been unable to cope with the balance between system performance and high concurrency,then microservice architectures have emerged as the times require.With lightweight services,efficient centralized management mode,and independent deployment methods,it has received unanimous praise in the industry.In order to maintain system availability and efficiency in high concurrency scenarios,it is particularly important to use better load balancing algorithms for microservice architecture design to fully leverage the performance of server nodes.This paper proposes the use of an improved weighted minimum number of connections algorithm to improve the load balancing capability of microservices systems,and the use of machine learning algorithm to predict the type of service requests.The prediction results are used to achieve pre-allocation of service requests using the improved load balancing algorithm.The paper conducts the following research based on the weighted minimum connection number algorithm:Firstly,based on the minimum connection number algorithm,the paper introduces system resource utilization as a load evaluation index,and proposes a dynamic adaptive load balancing algorithm based on the weighted minimum connection number.The algorithm periodically collects information on load metrics such as CPU,memory,disk IO,network bandwidth and utilization of requested connections of service nodes through message queues,and calculates the weight value of each service node using a load evaluation function constructed by the linear synthesis method.And the weights of each service node in the load balancer are adjusted periodically so that the load balancer can reasonably allocate requests according to the real-time performance of each service node in the cluster,so as to improve the utilization of server resources and fully utilize the performance of the server.Then,the paper uses the LSTM model modeling method in machine learning to predict the type of service requests,and combines the improved load balancing algorithm to pre allocate service requests.The results of the pre-allocation is recorded in the collection,and when the actual requests arrives,they will allocate directly based on the pre-allocation records in the collection,thus quickly hitting the service node in order to improve the allocation efficiency of the whole system.Finally,the paper designs and builds a new system architecture based on the microservice architecture,uses Nginx and gateway clusters to achieve a highly available gateway,and uses JMeter software to perform stress testing on service requests.Through experimental comparisons between the minimum connection count algorithm,the weighted minimum connection count algorithm,and the improved algorithm proposed in this article,the results show that the improved load balancing algorithm in this paper exhibits better node load balancing,faster response time,and higher throughput.