Storing incoming work dir on ramdisk

[Julian Field]

At 16:20 02/01/2003, you wrote:
>On Thu, 2 Jan 2003, Jeff A. Earickson wrote:
> > Julian,
> >    Sendmail has *always* been the performance pig and bottleneck
> > on my mail-server.  While MailScanner 4.x is great, the performance
> > of MS 3.x was sufficiently fast to outrun sendmail on my system
> > (a dual-CPU Sun E220R), always.
>
>Interesting.  Poor old sendmail always seems to have blame heaped upon it.
>So I'm going to try to defend it (just a little at least).
>
>With MS 3.x we (university with 100,000 messages/day) found MS (and its
>environment), not sendmail, to be the bottleneck.  This may have been
>because we did ORDB checks from MS, and the apparent MS slowness was
>actually DNS/ORDB latency.

Yes, that's what I always found too.

>Certainly the migration to MS 4.x, with its ability for multiple, parallel
>MS processes has helped matters enormously.
>
>For our site MS 4.x/sendmail-8.11 on Sun Ultra-10/Solaris-8 copes
>adequately.  The critical difference was upgrading MS from 3.x to 4.x.
>(But we are now moving to new dual-Intel Redhat, which seems even more
>comfortable.  Still sendmail!)

I still run sendmail here too. Our servers cope happily with the mail load
we give them, so I see no point in doing anything like trying to replace
our (complicated) sendmail setup with anything else. We have been running a
similar (in function) system since at least 1988 and it works just fine.
The amusing thing is that is operates in a very similar way to the way a
cluster of Exchange servers would behave (but has decent centralised
management, etc).

I have no plans to implement any production Exim servers here, but I was
quite surprised with the test results. Next thing to try is on a much
slower machine to see how much it is CPU-dependent.
--
Julian Field
www.MailScanner.info
MailScanner thanks transtec Computers for their support