Improving Proxy Cache Performance by Domain-Behavior Classification and On-Demand Caching

To increase the hit ratio on proxy and reduce the access latency of clients, different proxy caching techniques are designed to predict clients` surfing behavior for fetching their request pages ahead. However, these techniques have two deficiencies. First, the prediction is based on the history of clients` references, but these historical data are not always credible due to the variability of clients` behavior. Second, although these techniquszxes can achieve high hit ratios on proxy, their total traffic loads on network are high. In this study, we propose a novel predictive caching method called ODBC (On-demand Domain-Behavior Classification). The ODBC method first follows Pareto`s 80/20 law to clean data. It applies the concepts of entropy and sliding window to identify the exploratory requests and removes them when making predictions. Then, it tags popular pages and let those pages stay in the proxy longer than the normal ones. Experiments on real traces show that ODBC can improve not only the proxy hit ratio but also the network traffic loads.