Y.K. Liu, L.M. Cheng, and L.L. Cheng (PRC)
Neural Network, Load-Balancing, Fuzzy, Competitive Learning, Caching, Clustering Techniques, Online Learning The key of high performance in dispatcher-based approach is the distributing algorithm of the load balancer. It centralizes the servers’ information and gives the best solution for the incoming requests. Fig.1: A common topology of web site with multiple servers. All the client requests will be collected and forwarded to the selected backend servers. And the server responses to t
Traditional neural network scheduling techniques can to improve request cache hit rate of web servers; however, it cannot provide a good performance real web site because it cannot balance the server workload fairly. Here, we propose a fuzzy neural network technique by feeding back the real-time system usage with an updating mapping rules based on different requested objects categorized into different servers groups with different cache size and according to their input frequency to enhance the cache hitting rate of scheduling, simulation result shows that the proposed technique keeps 92% to 99% cache hit rate and in parallel finely balances backend server resource usage. The common architecture used in a multiple server web site is shown in Fig.1. The requests from client are received by a load-balancer and redistributed to different servers [3].
Important Links:
Go Back
