Research And Implementation Of A Load Balancing System Based On Web Server Clusters

With the rapid development of the information, the Internet plays an indispensable role in people’s work and life. The domestic and international Internet environment shows three aspects:the first one is a large base of Internet users rising trend; the second one is the huge number of network access; the last one is the business activities attached to large-scale on the Internet tending widespread. In order to provide the best services web, Create web server cluster is a good idea, and the core of web server cluster is load balancing scheduling strategy.The purpose of this study is that the Web server cluster shorten the response time of requests and still well provide high availability and scalable services in the environment as has been said above. So the combination of the three load balancing strategies that is based on Dynamic Feedback, Artificial Neural Network and Genetic Algorithm is proposed, and a load-balancing system based on the new scheduling strategy is designed and implemented. The main contents are as follows:Firstly, the research is based on the dynamic information feedback strategy. For the periodic collection of information on each node, using reliable factor clusters compre-hensive average index which is computed, and distribution ratio is corrected based on the index for request distribution.Secondly, this paper studies the applications of artificial neural network based on the dynamic feedback strategy, describes the design of neural network and simplifies the nerve input indicators, uses the CPU utilization, memory utilization, disk read and write rates, requesting a connection rate as four indicators. Uses training dynamic learning rate based on dynamic information data in order to artificial neural network training.Thirdly, this paper studies the application of genetic algorithm based on dynamic scheduling strategy. Global load balancing is based on cluster nodes the objective function. The initial value is artificial neural network output to weight so as to achieve the effect of weight correction, and get the global optimal weight distribution strategy.Fourthly, this paper describes the load balancing system’s design and imple-mentation process in detail which is. based on the combination of the three strategies as has been said above and based on the central processing model, namely receiving, distribution, classification, strategy, information gathering, scheduling and monitoring module design and implementation. Strategy module uses single-mode to design and joone and jgap to realize; Information Module uses sigar to realize; Monitor uses MVC pattern to design combined with Ajax and jFreet technology.Finally, this paper builds a test environment which uses Jmeter simulates multi-user concurrency testing, performance and load balancing system stress test. The conclusion is that in the experimental environment of this article, the proposed strategy is superior to rotation scheduling policy. These studies on the cluster system have certain reference value and significance.