We study the interconnection of Web document caches via ATM and IP networks. In this perspective, a document query travels trough several caches before reaching the document home server. If one intermediate cache has the queried document, it immediately sends it back and the query ends. There are several challenging problems to be solved. One is how to organise those caches : a hierarchical struture has a smaller latency than a flat one, on the other hand it introduces traffic bottlenecks. A second challenge is how to reliably transport data between caches : TCP, the Web transport protocol, limits the amount of unacknowledged data in transit and this limit must be small in IP compared to ATM. Therefore, traffic control between IP and ATM is critical to achieve good performances. Finally, typical drawbacks of caching like staled documents, false statistics and security must be addressed in this context.
We have installed a proxy HTTP server on a machine connected both to our ATM network and to the Internet over an ethernet connection.
To see the effect of the cache try to download the following file twice (this demo only makes sense if you have an ATM connection to our HTTP proxy):
Wolf cubs, 59k, compressed image, from a server in the USA.
Click here to remove the file from the cache of lrcsun5.