Network Computingwww.networkcomputing.com

The theory behind Web caching is simple: When users on your network request Web objects, the cache makes the request on behalf of the client to the origin server. In turn, the origin server sends the requested Web objects to the cache along with header information, such as whether the object can be cached and for how long.

If the content can be cached, the Web cache stores it on your LAN for subsequent requests. When the new requests come in, the Web cache can glean the age and accuracy of the cached content based on the original header information, and check with the origin server to determine if the older Web objects have changed since prior requests. The entire process must occur quickly to prevent end users from experiencing a noticeable delay.

Worldwide, caching has caught on--estimates of cacheable content online are now near 80 percent--but that doesn't translate directly into bandwidth savings. Members of the Web Polygraph team (polygraph. ircache.net) use a 55 percent hit ratio when they run caching bake-off performance tests. That number is derived from real-life caches that reveal hit ratios varying from 30 percent to 70 percent. Caching vendors' hit ratio claims are occasionally higher than these conservative estimates because some do not add prefetched documents into their hit ratio calculations.

Clearly, Web caching is a practical, efficient concept, but before you shell out big bucks to implement it on your network, there are some issues you'll need to address. How will this new technology affect your network configuration? How do you address a new point of failure? Can you expect the same trouble-free Web surfing that you already experience? Are other value-added services available in conjunction with Web caching?

The Issues
Two typical caching configurations are the familiar cache-enabled proxy and the transparent cache. The former is similar to a normal proxy server except that Web content is stored on a disk or RAM store. Other content often is proxied as well and sometimes cached for future requests. Most proxy caches also cache FTP, Gopher and NNTP (Network News Transfer Protocol) traffic. All proxy-based technologies are disadvantaged in that they require each client's configuration to be modified to receive forwarded content requests. While in the past this meant configuring each client individually, the advent of the PAC (Proxy Automated Configuration) URL has eased the burden.