.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <[EMAIL PROTECTED]> User-Agent: Mutt/1.4i
Quoting Clifton Royston ([EMAIL PROTECTED]): > On Mon, Feb 10, 2003 at 01:05:22AM +0100, Cliff Sarginson wrote: > ... > > principle .. "the mail must get through". > > Then application of a little logic yields the result that Internet > email is not of any use, because the Internet email system as a whole, > including the SMTP protocol, is not built on "the mail must get > through." > > Good admins treat correct delivery of mail (or non-delivery notices) > as an extremely high priority, and so does well-written MTA software. > However, there is a lot of mail system software in use on the Internet > that does not guarantee this and there are a lot of admins of Internet > systems who really don't care that much. For example, some MTAs may > acknowledge receipt of mail to the sending client before it is actually > written out to disk, Which makes it not SMTP, but "SMTP like". A client had a term for a vendor whose software we were replacing: They are "RFC aware". data MUST be written to non-volatile storage before being accepted. > and some admins may deal with some kinds of mail > system problems by deleting everything out of their mail queue in the > course of starting fresh. That makes the Internet as a whole unreliable > as a mail delivery system. And depending on the problem (massive disk/raid failures), yes that happens. UDP is, by definition unreliable. Yet for years we've used NFS and SNMP over UDP without an issue. The NFS protocol deals with it at a higher level. When you get an ACK from a user that she got the message, then you can count on her having gotten it. DSN's are handy in their failures, as pointed out. Knowing that a message is still in queue after 4 hours is handy for lots of mail. Lots of these features came about after X400 came in with its features. (while really really slow, you could find out where the message was). We don't want to get rid of the lightweightedness of SMTP and the rest of store-and-forward that made Usenet and the internet's subnetworks work. Users and admins will not "go for" having their POP server indicate that they got the mail (hell, fetchmail doesn't mean I've seen it, it just means that something got it from the spool). If you want true acknowledgement it must be dealt with at the layer higher than the transport layer (SMTP/POP in this case). That leaves us with the user, right now. Now, back to qpopper discussions
