| The explosion in usage of the World Wide Web has increased the load on servers and has made the network prone to serious congestion and User Perceived Latency (UPL) problems. To overcome such problems, web caching technologies, which are adopted on memory systems, have been introduced to the network.; In this thesis, web cache pre-fetching on the server side and re-fetching for cache replacement policy are studied. In the pre-fetching scheme, a prediction algorithm based on client request probability is proposed and evaluated. In the re-fetching scheme, a new adaptive re-fetching algorithm is introduced to shift dynamically among the existing policies based on their relative performance. Both algorithms develop the idea of machine learning in caching systems. On the client side, two client request models are introduced. These two models are employed to track the client actions based on probability algorithms and to predict the clients next action. In addition, web cache systems that implement adaptive caching by pre-fetching and re-fetching are simulated using Java programs. Results are measured by hit rates. Analyses of the results show that our new algorithms improve significantly the performance of the caching system.; Keywords. Web Cache, Pre-fetching, Re-fetching, Machine Learning, Hit Rate, Adaptive, Matrix, Replacement Policy, Latency. |
