Automating selection of an appropriate cache will require that nameservers provide geographically oriented responses relative to the source of domain name server queries. That is, a modified Domain Name Service (DNS) should resolve an IP address destined for the ``virtual'' server (NCSA's in this case, although a similar architecture would apply to any cached file depositories, in fact any other distributed information service), and redirecting the source to the server/cache closest to the client. ``Closest'' may reflect metrics including but not limited to physical distance or number of hops from the query source.
We have already started work in creating a conversion table to map IP addresses, to the best of our ability, to geographic locations within the US; we plan to extend this to world zones. An error margin exists since not all real network locations are known, and network numbers may span large geographic areas. But such a methodology offers reasonably high probability of locating the information requestor, and thus conserving system resources by facilitating their use of a local facility.
As with cached document timeouts discussed in section 4.2, appropriate DNS timeouts for web caching may require empirical evaluation. Servers that offer real-time traffic or stock market reports may require different timeouts than servers of Shakespeare plays or Biblical verse, which presumably do not change as frequently. Integrating such intelligence about the web, and the ability to gather it, into client and management software can allow dynamic selection of optimal servers based on roundtrip time, server load, policy, or other metrics.