Research On Technologies Of Web Applications Load Balancing Based On Dynamic Feedback Theory

Due to the rapid development of Internet technology, Web application server cluster has been widely used in e-commerce, etc. Load balancing is an effective solution for server cluster to improve the performance of the servers. How to distribute the huge incoming work flows reasonably in order to gain a better processing capacity, has been an important topic in the research of industry. In view of several features of Web application in the current Internet environment such as logical complexity, service diversification, user priority, and high load capacity, etc, the research focuses on the rationality and pertinence of load scheduling, aims to achieve better load balancing. Based on the analysis of some major problems in load balancing, the research of load balancing is based on the dynamic feedback theory, and it conciders the characteristics of current Web application, focusing mainly on the following work.In view of the deficiencies of previous dynamic load balancing strategy in current Web application environment which includes the timeliness of load information collection and the accuracy of load assessment, a type of load feedback and evaluation mechanism based on double periodic method is proposed, which introduces a feedback control theory, and predicts system’s average load in a relatively long period; defines load distance formula, makes each node estimate itselves load in a relatively short period for allowing light load nodes feedbacking load information selectively, participating in schelduing. The experiments show the rationality of load assessment by the double periodic feedback method, and it proved that the scheme based on double periodic feedback mode works better than than the previous single periodic feedback strategy did.An improved dynamic load balancing strategy based on services differentiation is proposed. In view of the service type diversification of Web application, a type of differentiation on user requests and nodes in the resourse level is proposed, through the grasp of adaptive correlation between them, which achieves a better targeted scheduling. In view of some characteristics such as user priorities and work sizes, a M/M/1differentiated services model is applied to queue the user requests, so as to reduce the average waiting time and ensure the quality of service. Concidering the high traffic characteristics of system, a certen randomness is introduced in the selection of nodes during the scheduling, so that the load balancer does not need to frequently update node’s load information in each short periodic, but make random scheduling by the relative static probability of nodes, and then according to the definition and calculation of node’s load distance, uses the total probability formula to compute nodes’ probability of receiving requests in order to improve the scheduling efficiency of load balancer. At last the experiments prove that the strategy has achieved a better performance in reducting the response time and assurancing the quality of service.