The Research Of The Load Balancing Technology Based On BP Neural Networks

As long as the computer network technology develops rapidly, more and more peopleobtain information on Internet. The access amount brought by the rapidly increasing numberof the active users is a great challenge to Web Servers’ ability of concurrent processing. It isnecessary to have web server technology of hign performance, high scalability, highavailability to increase the throughput and reduce the response time. Load balance cluster is agood solution, the key of which is the load-balancing algorithm. This paper presents adynamic load-balancing algorithm based on BP Neural Network，which dispatch requestaccording to the predicting average time by BP and current processing number of request.BP Neural Network is a multilayer feedforward network trained by Back PropagationAlgorithm. It’s one of the most widely used Neural Network modles at present. BP NeuralNetwork after training can predict the time of request processing, without establishingmathematical models. This paper firstly introduces the related technology of load balance，andthen falls into four levels: computer hardware, operating system, Java virtual machine, middleware, to analyze the key performance counter which influences the server performance, Thenuse BP Neural Network to predict the time of request processing. this paper design aload-balancing algorithm based on predicted process time, and implement this algorithm bydeveloping Nginx modules and monitor agent on Tomcat server side. At last the paperperform a benchmark based on TPC-W, to prove that this algorithm can improve theutilization ratio of cluster, compared with Weighted Round Robin Schedule algorithm.