From: Matthias Andree (madt.e-technik.uni-dortmund.de)
Date: Mon Aug 01 2005 - 15:15:13 CDT

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Paul Hutchings <paul.hutchingsmira.co.uk> writes:

> I have a basic grasp of Linux and our current smtp relay is Postfix and
> Spamassassin running on a Suse 9.0 box.
>
> I'm planning on replacing the hardware, and since Suse seems within my
> limited admin skills I see no reason to make life difficult for myself, but
> I'd like to confirm that someone is at least running Postfix in a simple
> smtp relay in/out config on Suse 9.3 with no problems?

Well, some understanding of Postfix won't hurt - I'm compiling Postfix
on my own usually within an hour of setting up a SUSE box, unless it's a
nullclient, which is just being fed an /etc/sysconfig/{mail,postfix}
pair before SuSEconfig --module is run.

> Also I have a hardware question, very basic I think.
>
> The current box is an ageing Dell 350mhz PIII with 256mb RAM and a pair of
> 9gb 10K SCSI drives on a hardware raid card.

That alone, if it were fresh hardware, would be pretty decent.

I have, until recently, been maintaining a 233 MHz P-II (the older
Klamath variant) w/ 128 MB EDO RAM that was
Postfix/Courier-IMAP/WWW/FTP/NFS server and replaced it because of its
age, it was nicely keeping up with the load for approx. 20,000 (give or
take 5,000) messages/week.

> The potential replacement is a spare box we have sat around, Xeon 2.4ghz,
> 256mb RAM but IDE only.
>
> Presumably in a low throughput setting such as ours (20-30k messages a week)
> this will be a beneficial move because I figure a lot of our overhead is
> spent running spamassassin against messages, I suspect we are disk
> bound.

This isn't clear to me. If you're disk bound, you'd certainly want to
stick to 10,025/min SCSI drives, and you'd want to be even more loathe
to give up redundancy (RAID).

SpamAssassin is pretty much CPU-bound, unless you use the awfully slow
Bayes functionality.

The difficulty with IDE is, for reliable delivery you will have to turn
the caches off, and there's no TCQ to save you, except if you're going
SATA + NCQ or SATA-II and use VERY recent Linux kernels and hardware. I
don't know how other operating systems deal with write caches, but the
common recommendation is "off" for mail servers.

The safe combinations I know are listed at
<http://home.pages.de/~mandree/linux/kernel/safe-write-caches.html> - if
someone has comments about that document, I'd be happy to update it.

HTH,

--
Matthias Andree

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]