Research And Implementation Of Session Maintenance In Load Balanced Cluster Under Cloud Platform

At present,the cloud computing industry has developed explosively and has been used in various fields in china.This has led to a dramatic increase in the demand for computing resources on the cloud platform.In order to deal with this situation,using load balancing cluster technology is the best choice.Its highly available,highly scalable,low-cost features are also favored by the industry.However,in the web application field,because the http protocol itself is a stateless protocol,and e-commerce and other services need to maintain the session state,the session maintenance technology is used on the basis of the http protocol.The technology works well in a stand-alone environment.However,in a load balancing cluster environment,session maintenance technology can cause the logical confusion of web applications due to changes in service nodes before and after.Therefore,In order to promote load balancing clusters to the e-commerce field,how to maintain sessions in load balancing clusters must be resolved.Based on the research and analysis of the existing load balancing technology and session maintenance technology,this thesis proposes a session sharing model,which solves the problem of session maintenance in clusters,and maintains excellent performance under high concurrence.First of all,the thesis designs a session sharing model by analyzing the existing web cluster.The model is divided into three modules.The load balancing module designs a new load balancing algorithm,which is based on the request content classification,calculates the performance margin ratio of different request types on the node,and dynamically feedback to the load balancer,and the load balancer according to the feedback information scheduling request.The scheduler schedules requests based on the feedback load.At the same time,in order to avoid back-end node overload,a threshold triggering mechanism is designed.The mechanism uses double-threshold to refine the state of the node.When selecting nodes,a random forwarding mechanism based on distribution weight is adopted.It determines the forwarding target nodes according to the location of random numbers,avoids the association between the two distributions,and makes the task assignment more balanced.The session management module changes the traditional way of storing session data by using local files to multi-layer distributed storage architecture that combines local memory and remote redis clusters.And based on this architecture,a session management scheme is designed.The session storage module is based on the redis database,designing the data structure and the expiration time of the session storage,selecting the appropriate persistence mode,and deploying the redis distributed cluster by using the codis scheme.Then,the thesis specifically implements the above three sub-modules and integrates them into a cluster system.Finally,in the actual hardware and software environment,we deploy and configure the entire session sharing model system,use test tools to test the system's session maintenance and response time of different cluster solutions under high load,and verify the function of system session maintenance and better performance based on the test results.