[ On Mon, March 10, 1997 at 13:13:24 (+0000), Nigel Metheringham wrote: ]
> Subject: Re: Number of recipients
>
> Mailers which are trying to be efficient in connectivity requirements have
> to route all the addresses first which leads to an initial delay - ie
> increased latency. This latency can be substantially reduced by taking a
> middle path which is to group by domain only and then to route and deliver
> each domain separately - ie you ignore any connectivity efficiency gains
> that would be obtained by factor 2 above.
Personally I don't give a damn about what I'll call "reasonable"
latency. E-mail is a store-and-forward technology, and people who
attempt to imply otherwise are misleading. If I want immediate
communications with a person or the person's agent I'll use a protocol
more appropriate for direct communications, such as talk, irc, internet
phone, or whatever. I've made good use of the delivery priorities in
larger smail sites to manage local delivery of ``junk'' mail in special
nice'd queue runs at less regular intervals. Even for 1500 users who
get lots of junk mail, this makes a big difference on the load the
mailer puts on the system.
I do feel that it is important that latency not be unduly increased
because of poor routing or queueing algorithms and implementations.
This includes not being overly agressive at trying to ensure the maximum
number of messages are delivered during a given connection.
So, to bring this rambling back to how mailers handle large mailing
lists.... I think in general the the mechanisms used by zmailer are the
best I've seen, though I'm not an expert in these algorithms. I'm not
even sure I know how zmailer differs from exim or smail, for that
matter. Zmailer is definitely even still a bit hard on system
resources, and perhaps it works too hard, though my guess is its current
implementation can still be optimized further. Ultimately real-world
delivery stats are the measure, but only when communications capacity to
each end-point is also factored into the equation. It would also be
good to keep in mind the differences between local delivery vs. remote
delivery when looking at mailing lists...
--
Greg A. Woods
+1 416 443-1734 VE3TCP robohack!woods
Planix, Inc. <woods@???>; Secrets of the Weird <woods@???>
