Cache Selection

A single cache, of course, won't do. There must be multiple caches of the same original content, possibly located around the world for popular Web and video streaming sites. An interesting technical problem is how to determine the optimum cache from which content should be retrieved.

Consider the case of a master Web site located in Silicon Valley. Caches are installed on the East Coast, Europe, and Japan. Now consider the case of a user in Australia. Which cache should provide the information?

Intuitively, the nearest cache as measured in geographical terms is preferred. In this case, the nearest cache geographically would be Japan. But there may be instances when that cache is down, or when the link between Australia and Japan is heavily used, or when the link speeds between Australia and Europe are faster than to the other caches, or when the number of administrative domains (BGP hops) is greater between Japan and Australia.

The general approach to the logistics of cache selection is to modify the root server of the Domain Name System (DNS) of the domain being interrogated. The modification is to give the DNS the responsibility of maintaining the IP address of each cache. When the user requests the site, a real-time calculation occurs of the distance between the end user and each cache. This distance refers to a metric that enables the DNS to return the best cache.

The selection of metrics, however, is more art than science. It could include link utilization, the number of router hops, the number of administrative domains, link speeds, or other measures. So, innovation is under way to develop new algorithms and modifications to DNS. Among the current products are Cisco's Distributed Director and Akamai Networks, mentioned previously.