- From: <bede@scotty.mitre.org>
- Date: Fri, 10 Feb 1995 22:26:18 -0500
- To: http-wg@cuckoo.hpl.hp.com
- Cc: uri@bunyip.com, connolly@hal.com
Date: Fri, 10 Feb 95 09:07:20 CST From: connolly@hal.com (Dan Connolly) [ . . . ] But as another solution, it seems that something like the MX record facility for finding SMTP servers would work well for HTTP servers. This "HX record" would map a domain name to one or more domain/port pairs or [ . . . ] The basic problem is that at the moment of accepting a client's connection, the server has in hand the IP address of the client but not a clue about either the IP address or domain to which the client believes it is connecting. Hence, an "HX" won't do the server any good for making a decision about what to do next. The notion of something like an "HX" record analogous to an MX record might make some sense despite this, since some sites will want to support off-site or redundant HTTP servers. But a new "HX" type is NOT a good idea because it requires formally extending the DNS spec. However, you can easily implement the exact equivalent of an "HX" RR type by using Hesiod's TXT type RR, which allows for application-specific syntax and semantics in its text portion. For example, you might have this pair of records for "bigsite" and "smallsite": www.bigsite.com. HS TXT "http://provider.net/bigsite http://other.provider/bigsite" www.smallsite.com. HS TXT "http://provider.net/smallsite" The syntax of the text string is simply a list of URLs: a client receiving this RR would try the first URL, then the second, and so forth, when searching for the site's "Home Page". For example, given "http://www.bigsite.com/", after retrieving the Hesiod record for "www.bigsite.com" the WWW client would start out by trying to GET "http://provider.net/bigsite" and then "http://other.provider/bigsite" if the first attempt failed. As a default, you could obviously use the same rule used for MX records: absent our "HX" record, the HTTP client should attempt to connect directly to the domain mentioned in the URL (i.e. "www.bigsite.com"). On the downside, unlike the case an MX record, you can't discover this HS class TXT record for "www.domain" by simply asking your local nameserver for class "IN", type "ANY" records for the domain. You have to make a separate query using the "HS" class to get the record. Caching at the local nameserver should reduce the impact. The fact that DNS zones are (in theory) always replicated means that this record will (in theory) be redundantly available to all clients on the net. - Bede McCall <bede@mitre.org> The MITRE Corporation Tel: (617) 271-2839 Bedford, Massachusetts FAX: (617) 271-2423
Received on Friday, 10 February 1995 22:47:53 UTC