On Fri, Jan 05, 2001 at 07:57:35AM -0800, Michael Cope wrote:
> This is the fallback MX host for all of our mail
> servers so any time an initial delivery fails at Earthlink it get's to
> this box. Overall we deliver about 2 million messages a day to 4 million
> recipieints.
Yes, that's what we shift. What's your network output? Are you on
gigabit?
Sometimes, smaller servers will be better than one huge one...
> > RBL and DUL lists are good. Otherwise, you are back to clever filter
> > files and humans.
>
> We've got this on the front ends, I suppose this is the way to really
> attack the problem. We also have manual tools for despamming the fallback
> queues, but this is very time intensive.
There are others methodes, as Nigel said... strange that we use just
those, isn't it? ;>
> > > Any other experiences people would like to share about Exim in a
> > > large volume environment is welcome.
> >
> > Sure, that's what we do...
> >
> > What exactly do you want to know?
>
> Mostly, does it break and how? Do people upstream report any
> problems. How much extra scripting and babysitting did/do you have to
> do. I'm not particularily interested in local delivery since we have our
> own software for that.
Break? when there is an ugly bug in exim ;> we found a few here.
However, Phil has always solved those in less than a day.
Disk IO is your bottle neck, and using a RAM disk does improve things by
a factor of about 4.
Best way is to do MRTG-like graphs of CPU/MEM/IO etc... and see what the
trends are.
--
Dr Yann Golanski Senior Developer
Please use PGP: http://www.kierun.org/pgp/key-planet
