At 14:52 14/09/2005, Alex van den Bogaerdt wrote:
On Wed, Sep 14, 2005 at 02:16:06PM +0100, Paul Smith wrote:
> >I'm saying that if the average time for a transaction (which is under the
> >control of the receiving MTA) increases, there will be a definite
> >which will be most keenly felt by high-volume sites.
> So, RFC 2821 needs to be adjusted so that the post-DATA timeout is set to
> less than 30 seconds? If that's what you're saying is necessary, then
> that's what the standard should say... Currently it says 10 minutes!
The __AVERAGE__ time is important here. If ALL processes need to wait
600 seconds, it is the worst case scenario. If on the other hand only
a few need to wait 600 seconds and most others finish in 1 second, it
is easily seen that the average could very well be under 30 seconds.
If you can reject in the RCPT TO stage (virtually no time needed) you
need not process the message through spamassassin, amavis, SPF, whatever.
This saves time so the AVERAGE time needed is less. And this is just
an example, there will be more possibilities to save time.
Agreed. The problem is that lots of SMTP clients DON'T allow the slow
messages, they just timeout after 30-60 seconds.
I've just looked at our server logs, and most messages get processed
through our server (including virus & spam filtering) in 1-3 seconds.
However, a few have big attachments (which take extra time in the virus
scanner) and/or lots of text including lots of URLs (which take extra time
in the spam filter, some have 100+ different URLs, our spam filter checks
all these out), these may take 75-90 seconds to be processed - but the
client times out (as it shouldn't) , and THEN resends the message (as it
You're going to have trouble making me believe that it's more efficient for
the client to send the 10MB message, wait 60 seconds, timeout, then 5
minutes later send the entire 10MB message again, and again, and again,
rather than send the 10 MB message once, and then wait up to 600 seconds
(in actuality it would only need to wait 90 seconds or so) and then finish
with the message once and for all.
The AVERAGE time for a message to be processed by our mail server
(including virus scanning & spam filtering) is probably 10 seconds or less,
but the peak time can be up to 2 minutes or so. This seems to be well
within what people here and the standard thinks are acceptable, but we have
problems because of the very short timeouts in the clients.
Hence the problem
Hence the need for a solution :)
Paul VPOP3 - Internet Email Server/Gateway