woods@??? (Greg A. Woods) writes:
> [ On Monday, April 8, 2002 at 19:49:45 (+0200), Florian Weimer wrote: ]
>> Subject: Re: [Exim] Mailing lists with a twist
>>
>> I fear that I might generate the messages much faster than they are
>> sent out by the MTA, and they are unnecessarily stored in the queue
>> (when it would have been better to delay the message generation a
>> bit).
>
> Does that matter?
I'm not sure. Injecting the messages over SMTP has a considerable
advantage: a stable interface, and no need to upgrade the injecting
software each time the MTA is upgraded.
> I would say you _want_ this to happen anyway -- get all the messages
> into the mailer's queue ASAP and then let it spew them out at the speed
> they can be sent over your available bandwidth and the speed the remote
> servers can accept them, using the amount of concurrency your local
> server and your available bandwidth can manage (but hopefully not using
> an MTA that will open [too many] multiple connections to each individual
> remote server).
Well, I'm currently bypassing the central mail relay, after some
(artificial) tests showed that it might not survive the load. ;-)
You are probably correct that storing all the messages in the queue
first wouldn't be too bad in the end. After all, our messages are
rather short (probably 8K or so in the average), and even tens of
thousands of them wouldn't take significant amount of disk space.
> The MLM should open a single SMTP connection to the local relay mailer
> which will deliver these messages and send each message separated by an
> RSET command.
When I do this, shall I implement pipelining in the MLM SMTP client?
Or this this a wasted effort if the relay MTA is Exim?
> I should probably reiterate that what you're doing is "bad" and
> unneighbourly.
I few years ago, I had the same attitude, but I'm no longer sure if
it's the right approach. It allows us to unsubscribe bouncing and
autoreplying addresses immediately, thus saving bandwidth. In
addition, network bandwidth is cheap, compared to the time required to
manually process bounce messages, or even to write complaints about
bounces without *any* information.
--
Florian Weimer Weimer@???
University of Stuttgart http://CERT.Uni-Stuttgart.DE/people/fw/
RUS-CERT +49-711-685-5973/fax +49-711-685-5898
