In the web, it is well known, that ``popular documents are very popular'' [9]. Thus, for each server, a small set of its files amounts for the largest percentage of web requests to that server. We call this set of documents Top-10, for the most popular documents of a server. Only documents that are members of the Top-10 are considered for prefetching by the clients. The actual number of documents in the Top-10 is fine-tuned based on client profiles which reflect the volume of requests initiated in the recent past by the client and the amount of disk space allocated for prefetched documents.
This idea of prefetching only popular items is not new: it has been followed for several years now by music shops. A significant percentage of a music store's stock contains the most popular LPs of each week. Both music store owners (proxies) and music store customers (users) make their purchases based on the Top-10. The actual purchases themselves determine next week's Top-10, which will determine future purchases and so on. The whole process of creating a Top-10 chart, distributing it to popular music magazines and channels, and making purchases based on the Top-10 of the current week, is being successfully used by millions of people each week all over the world.