Research On A LVS Cluster System Based On Cookie Session Maintenance

This thesis uses the cluster technology to build digital city management platform.For the purpose of the study, it uses the four layered load balancing technology IPVS ofopen source software the Linux Virtual Server project as the cluster of customizedsolutions. Due to the complex operation scene of the load balance together with IPVSmaintainer energy is limited, the IPVS still exists many defects in the application ofneeding to keep the session, especially in the high-speed proxy environment. Becausethe IPVS adopts the session-persisting solution based on the source IP, there may existsession lost and severe load imbalance problem. Based on the above problems, thispaper studies the IPVS load balancing implementation, puts forward the solution of theproblem and analyzes the performance of cluster server.The main research work of the paper is as follows:①This thesis studies the cluster system combined with the needs of digital citymanagement platform. After studied the cluster of the traditional sharedsession-maintenance, this paper puts forward the solution which integratessession-maintenancefunction to load balancer. But IPVS is based on the source IPsession-maintenancesolution, it may exist session lost and severe load imbalanceproblem in the high-speed proxy environment. So, it needs to improve the IPVS.②In order to research IPVS implementation principle and system structure, itshould be firstly analysis of the TCP/IP protocol principle, the Linux kernel of networkpacket processing, the realization mechanism of the Netfilter, the IPVS Netfilter basedon IP packet handling, etc. Secondly, it should analyze the source and implementationdetails of the IPVS, put forward the load balancing algorithm based on Cookie SessionMaintenance, carry on feasibility analysis and algorithm design and achieve thealgorithm logic finally.③In view of the digital city management platform architecture of cluster system,this paper establishes the performance analysis model and corresponding simplifiedmodel based on queuing theory, and uses probability statistics method to get theanalytical expression of the system average response time to analyze the influence ofvarious performance to analytical expression.④By constructing environment for experiments, experiments have been carriedout to verify the correctness and effectiveness of the algorithm proposed in this paper, and the performance test and comparative analysis of two alternative solutions in thecluster system for the digital city management platform. Experiments show that theproposed algorithm can effectively solve the problem which there has a chance to leadto session lost and severe load imbalance in the high-speed proxy environment, and onthe cluster performance is better than shared session clustering.