28000 mails in the queue and you're using the chkuser patch? My guess is
that over 27000 of those are spam double-bounces waiting to be delivered. I
might suggest setting up another machine to be a smart smtp host to handle
outgoing mail. Trying to resend all that mail takes non-trivial resources.
You might also consider lowering your queuelifetime, which will
double-bounce things quicker that will likely just sit in the queue until
they expire.
--Doug
Jeff Koch wrote:
> Hi Jeremy:
>
> Thanks for replying.
>
> We're using spamc in a maildrop filter called for each pop account
> (via qmailadmin's spam enable setting). User SA prefs are maintained
> by. MySQL
>
> Good suggestion on the catch-alls - we'll do a scan for accounts with
> many emails. We do have some with 10 to 15,000 emails. ReiserFS was
> not an option on RH8.0 unfortunately.
>
> We've got 28,000 emails in the queue - not preprocessed is zero.
> - concurrencyincoming = 30
> - debugging is turned on for qmail-scan (seems to be the only way to
> generate a log of all messages processed) but the log is closed and
> zipped each night.
> - smtpd log shows we're using about 15 of the 30 incoming connections.
> - softlimit in /var/qmail/supervise/qmail-smtpd/run is 30MB
> - vpopmail is configured for MySQL auth with all domains in one table.
> - we use mfcheck and the checkuser patch (mysql) to drop questionable
> connections.
>
> I don't know what you mean by 'conf-split' or big-{ext-}.
>
>
> At 02:01 PM 2/27/2004, you wrote:
>> On Fri, 2004-02-27 at 12:53, Jeff Koch wrote:
>>> Does anybody have any suggestion for improving the performance of a
>>> mailserver runnning
>>> qmail/vpopmail/qmailadmin/qmail-scanner/spamassassin?
>>
>> are you using spamd and having qmail-scanner call spamc? or is it
>> calling spamassassin. if the latter, change it to spamd/spamc model,
>> much more efficient.
>>
>>> We're using a 2.4Ghz P4 with 1GB RAM and a 40GB SCSI drive. The
>>> operating system is a basic RH8.0 install with the ext3 journalling
>>> file system. We're handling about 50K messages/day and seem to
>>> exhausting the capabilities of the server. During peak periods CPU
>>> idle time reaches 0% and load averages will exceed 10.0
>>
>> with ext3 you have to be careful about catchall accounts, and
>> accounts that nobody checks that get a lot of email. If there are
>> say, 25000 messages in an account and it gets 100 messages per hour,
>> that will bring your server performance down quite a bit.
>>
>> of course, using reiserfs on the mail store will fix that problem.
>> I've had over 150k emails in one account with reiserfs, and the
>> system wasn't even being phased as I pumped more mail at it (was
>> load testing for a customer)
>>
>>> We would welcome any suggestions or URL's that you could point to -
>>> particularly with respect to performance tuning that will work with
>>> qmail.
>>
>> what is the approximate queue load, and what is your conf-split
>> value?
>>
>> do you have big-{ext-,}todo?
>>
>> -Jeremy
>>
>> --
>> Jeremy Kitchen
>> Systems Administrator
>> [EMAIL PROTECTED]
>> Kitchen @ #qmail on EFNet - Join the party!
>> .....................
>> Inter7 Internet Technologies, Inc.
>> www.inter7.com
>> 866.528.3530 toll free
>> 847.492.0470 int'l
>> 847.492.0632 fax
>> GNUPG key ID: 93BDD6CE
>
> Best Regards,
>
> Jeff Koch
