Hot Spot Prediction Algorithm for Shared Web Caching System Using NN

As the population of Web grows, network traffic on the World Wide Web increases, which has a great impact on the network utilization. On WWW, there are innumerable objects, popular and unpopular of which the frequently requested object by users is called ´hot spot´. Often hot spot brings an excessive load to the cache server and original server, resulting in a swamped state in the system. In this paper, a hot spot prediction algorithm based on neural network has been suggested to solve the problems induced by hot spots. The hot spot to be requested in the near future will be prefetched into the proxy servers after predicting. Therefore, a faster responding to the users´ requests and a higher efficiency of the proxy server can be achieved. Hot spots can be obtained by analyzing the access logs file using the neural network method. A simulator has been developed by using the PERL language in order to validate the performance of the suggested algorithm, through which the hit rate improves and the requests among the shared proxy servers are well load-balanced.