| The rapidly increasing usage of the World Wide Web has resulted in an increase in load on the web servers, and the amount of data transferred over the network. In order to minimize network traffic, web-caching techniques are used. To complement web-caching and improve users' experience, the concept of pre-fetching has been introduced. Various pre-fetching techniques and models have been proposed in the recent years, however, each of these models have their limitations and advantages.;The proposed pre-fetching models and a web-cache simulator have, been implemented in Java to measure the efficiency of the models. In conclusion, no single model turns out to be the best in all situations. For caches sizes between 40/0-25%, the multi-dimensional matrix models achieve satisfactory results; whereas, for cache sizes as small as 2%, the two-dimensional single-prefetching model achieves better hit rates.;Keywords: Web Cache, Web-caching, Pre-Fetching, Multiple Pre-fetching, Hit Rate, Web Cache Simulator, User Perceived Latency (UPL), Matrix, Multi-Dimensional Matrices, Markov Tree;In this thesis, we have approached pre-fetching from two angles: simplicity and computational efficiency. We have built a Simple Web Object Relationship Model that is able to avoid complexity and the costs involved in black box models based on data mining, and other models that require frequent and intensive training. Inspired by the Markov Tree model for pre-fetching, we begin by building a simple tree structure for the Simple Web Object Relationship Model and then propose and build Multi-Dimensional Matrix models. From the best of my knowledge, the use of multi-dimensional matrices for pre-fetching models has been proposed for the first time. |
