Researches On Improvement Of Health Detection And Load Balancing In Spring Cloud Based On Load Forecasting

As one of the most popular microservices framework,Spring Cloud has received more and more attention and use.Spring Cloud complete health detection through heartbeat mechanism,which can only detect whether the service is running,instead of whether it can run successfully.At the same time,the load balancing algorithm in Spring Cloud simply considers some statistical variables and cannot provide an optimal balancing strategy in combination with the actual operation of the service node.In this thesis,load prediction is introduced to deal with the problems in health detection mechanism and balancing load algorithm in Spring Cloud.An improved load balancing algorithm is proposed based on the results of load prediction,in which the health detection mechanism is optimized by load prediction.The improved algorithm provides a reference optimization solution for building the Spring Cloud microservices framework,which has important practical significance.The main work presented in this thesis includes the following aspects:(1)Through the analysis of load characteristics and several load prediction methods,this thesis selects the time series model as the load prediction method,and gives the definition of integrated load,and selects the appropriate time series model and model parameters for the experimental environment.(2)In view of the shortcomings of Spring Cloud health detection mechanism,this thesis combines the health check ideas with some optimization schemes of other existing frameworks,integrates the Spring Boot monitoring component with the original heartbeat mechanism,to monitor the normal operation of the service.External resources are used to judge whether the service instance is normally available.At the same time,with the idea of healthy grading,the service load forecasting situation is used as the grading standard to implement the grading strategy of the health service,which provides the data support of the subsequent load balancing algorithm.(3)In view of the shortcomings of the existing load balancing strategy in Spring Cloud,starting with the good ideas in the Spring Cloud load balancing algorithm,thisthesis propose a dynamic weighted load balancing algorithm DWRR(Dynamic Weighted Random Rule)which based on the service instance prediction load obtained by the health detection.It combines the filtering ideas and weight random ideas in Spring Cloud's original load balancing algorithm,introduces the concept of strong service providers,and integrates the number of instance connections,response time statistics,the service health grading and load prediction values to dynamically adjust weights and select instances.The results of experiments show that the algorithm can effectively improve the comprehensive utilization of the system in the heterogeneous service cluster environment.