You are right
I missed out on some critical data:
The box is just playing a relay domains are lsearched in a flat file (only 15 entries) en pushed to the appropriate internal AV boxes.
The OS is a standard redhat linux 8.0 with an ext3 filesystem but the system seems to be more CPU bound than IO bound since it is forking itself to dead.
The standard acl is generating tons of dns traffic as it tries to verify the addresses.
Apparently this is not cached ... or I missed a setting here...
If I send 1000 messages to the same destination domain it does an mx lookup of the sender and recipient domains for each of them.
I have tried both with a local caching nameserver (on the box) and one on the same segment without a significant difference
The box is handling a mix of in/outbound mails which includes some mailings to customers (batches of 20.000 and more) and the normal corporate mail stuff/spam.
Indeed I'll try to fiddle with the load/queue settings to keep system load on a reasonable level but I was just surprised/afraid that this will keep my max concurrent smtp sessions to something around 50 and that performance will drop further when adding more checks (e.g. rbl, local antispam rules) and some rewrites
50 concurrent sessions is not giving the throughput of 120K/mesg.
Thanks for your tips.
Ronny
On Thu, Apr 03, 2003 at 11:41:55PM +0200, Nico Erfurth wrote:
> Ronny Vaningh wrote:
> > Hi
> >
> > I want to persuade my boss to migrate our mailsystems to exim.
> >
> > I tried to stress test my mailserver running 4.12.
> >
> > The hardware is a dual P3 XEON700 with 1 GB of RAM
> > RAID1 message spool and separate RAID1 system/log partition on 10Krpm
> > scsi disks.
> >
>
> How you handle your domains, what kind of database do you use?
> Do you run a local caching-daemon?
> What OS do you run?
> What filesystem do you use?
> Do you want to do normal mail-delivery or mainly mailinglists/newsletters?
> Do you test remote or local delivery?
>
> The "problem" is that exim tries to send out mail as soon as possible,
> in fact it starts the delivery in the moment it sends back the 2xx OK
> for the DATA.
>
> A good way to cope with this is to set queue_only_load to a reasonable
> value and use a queue_runner_max of 20 or so, then set you queue-runner
> to 1 or 2 minutes. That will slow down the delivery a bit, but it also
> will regulate the amount of running exim-processes. Tuning these values
> should help to get the best out of the machine.
>
> Tell us more about your current setup/config and we can tell you more
> about what to do.
>
> Nico
