Received: from localhost (daemon@localhost) by CS.UTK.EDU with SMTP (cf v2.9s-UTK) id MAA26560; Sat, 30 Jan 1999 12:59:39 -0500 (EST) Received: by cs.cs.utk.edu (bulk_mailer v1.11); Sat, 30 Jan 1999 12:58:32 -0500 Received: by CS.UTK.EDU (cf v2.9s-UTK) id MAA26428; Sat, 30 Jan 1999 12:58:30 -0500 (EST) Received: from a4.jck.com ([206.99.215.40]) by CS.UTK.EDU with ESMTP (cf v2.9s-UTK) id MAA26373; Sat, 30 Jan 1999 12:58:18 -0500 (EST) Received: from p6.jck.com ("port 1209"@[206.99.215.34]) by a4.jck.com (PMDF V5.1-11 #C3296) with SMTP id <0F6D00F09WL4SO@a4.jck.com> for drums@cs.utk.edu; Sat, 30 Jan 1999 12:58:17 -0500 (EST) Date: Sat, 30 Jan 1999 12:58:05 -0500 (Eastern Standard Time) From: John C Klensin Subject: stripped-down client configuration (Address, MX RRs, A RR-only?) To: drums@cs.utk.edu Message-id: MIME-version: 1.0 X-Mailer: Simeon for Win32 Version 4.1.5 Build (42) Content-type: TEXT/PLAIN; CHARSET=US-ASCII Priority: NORMAL X-Authentication: none List-Unsubscribe: Open question... When one has a stripped-down SMTP client (e.g., one associated with a POP3 or IMAP client), that client is usually configured to forward mail to a "real" SMTP server for relaying. There are three theories about how that configuration should be done and interpreted by the client (I've seen all three at one time or another): (i) Client configuration should use an IP address Pro: no ambiguity about what host is being referenced, saves DNS lookup time. Con: coded-in IP addresses, even in config files, are a bad idea, especially with IPv6 coming. Addresses reduce flexibility. (ii) Client configuration should use a domain name, which should then be processed through MX resolution, etc. Pro: very flexible behavior, consistent with what real SMTP servers do, flexibility should be under the control of the server operator. Used by at least some ISPs to provide alternate relays when some are busy. Con: Lower performance, client should be able to control everything, regardless of preferences of server operators and zone admins. (iii) Client configuration should use a domain name, but it should be interpreted as a reference to an A RR, i.e., no MX resolution. Pro: More control by client, higher performance than MX resolution, reduction in risk of loops from badly configured list of MX hosts. Con: A bit abnormal relative to how we handle mail; some names of interest may not have A RRs or the MX RR may point to places besides the host associated with the A RR. Of course, the first can be combined with either the second or third. Observation #1: if an IMAP or POP3 client (probably the most common cases of clients that need to be configured this way) is queuing up mail and then sending multiple messages to the relay in a single SMTP session (as currently suggested in 1123 and 821bis), the overheads here are per-session, rather than per-message, and may not be significant on a per-message basis (or at all). Observation #2: Regardless of what decision is taken by DRUMS on this subject, "dumb host can't support MX resolution" should not be an excuse: such hosts are already non-standard and non-conforming, per RFC 1123 and elsewhere. Internet hosts are required to support DNS resolution, which includes MX resolution. At the moment, 821bis leans strongly toward the second option, but may not do that clearly enough to leave no ambiguity. Anyone want to argue for the first or third approach? Or to propose text for any of the three? --john