Jacob Lundberg

CS 312 - USENIX Summary

Prefetching Hyperlinks

Getting web content before it is needed is not a new idea. It is used extensively in transit caching such as proxies and client-side caching such as the cache of a web browser. However, in a USENIX article by Dan Duchamp of AT&T Labs, a recent study in client prefetching showed a considerable improvement in traffic reduction and client latency. Because this fetching is done by individual clients, a much larger percentage of data may be prefetched. A client's cache is for that client only, so dynamic content based on cookies, for example, may be cached. The client communicates with servers that have been contacted recently or that it considers likely to be contacted and the servers tell the client what content their content has links to and the average click-through rates on that content. Then the client decides what content to prefetch based on what the server has told it. The results found by the research team included a 24.0 percent average reduction in wasted bandwidth and a 52.3 percent lower average client latency. This is not now a production system, and it is probably some time away from entering the mainstream, but it will likely be an excellent option for System Administrators and the Clients of their web services in the future.

Referenced USENIX article:

http://www.usenix.org/events/usits99/full_papers/duchamp/duchamp_html/index.html