Re: server redirects by client domain

Donald E. Eastlake 3rd (
Tue, 26 Dec 1995 15:19:06 -0500 (EST)

On Fri, 22 Dec 1995, Larry Masinter wrote:

> There are at least three proposals floating to use 'DNS' to change how
> references work:
> a) use DNS for finding URN resolution services
> (cf. various draft-ietf-uri-urn-* proposals).
> Among other things, URN resolution services would help you find mirror
> sites and replicas of the 'same' resource.

This sort of thing involves looking up URN's or URN fragments and
would probably best be done with a new DNS RR type.

> b) use DNS to indicate HTTP-NG vs HTTP support.
> Simon Spero mentioned this in his talk describing HTTP-NG, that DNS
> records might indicate the use of a completely different protocol for

Given that you are down to the level of trying to connect to a host,
you have to go to DNS to translate the host name into an IP address
anyway. There exists a defined RR type known as WKS for Well Known
Services. Looking up just the "A" address records for a host name
gets you one or more IP addresses with no additional information.
Looking up the WKS RR's for a host name gets you one or more records
each of which has an IP address plus a bit map of sockets less than
256 that host listens on (if the WKS's have been set up and maintained
for that name. Thus, using WKS or something like it, you could get
bit maps with bit 80 on in HTTP is supported and bit ?? a one if
HTTPng is supported. You could even have, say, four WKS records
stored under one (say name such that they
specify four different IP addresses of four different machines one
supporting no HTTP, one supporting current HTTP, one supporting only
HTTPng, and one supporting both flavors of HTTP. There are some added
complexities like having to fall back on type A records if no WKS
is present and WKS not being adapted for IPv6 addresses. Given
the rate of web evolution/deployment and the bad history of WKS RR
maintenance, it may work just as well to just try HTTPng and falls
back to HTTP if it fails, caching the info.

> c) use DNS for simple host replication/round robin to indicate
> multiple sites at different geographic (network topology) locations.

I'm not familiar with how it has been proposed to do this. It is not
clear to me that, with a reasonable amount of effort, you can do noticeably
better than a longest prefix match between your IP address(es) and those
of a server you with to hit.

> Whatever you think of these individually, doing them all with separate
> DNS entries seems like a bad idea.

"b" & "c" seem very similar but I don't know about "a" being the same.

Donald E. Eastlake 3rd +1 508-287-4877(tel)
318 Acton Street +1 508-371-7148(fax)
Carlisle, MA 01741 USA +1 703-620-4200(main office, Reston, VA)