MS Performance

[Tom Sevy]

Don't know what you use for Firewalling, but I have taken the top invalid
recipients and created a rule in our Firewalls to reject email to them.

----- Original Message -----
From: "Ken Anderson" <ka at PACIFIC.NET>
To: <MAILSCANNER at JISCMAIL.AC.UK>
Sent: Monday, July 14, 2003 12:30 PM
Subject: Re: MS Performance


We relay about 500k emails a day through 2 MS machines running redhat &
sendmail. Both machines are dual Xeon/1gb RAM/SCSI, RAID1
About 45% of the total incoming mail is spam, half of which is deleted.

With the exception of the following, it's a default rpm install:
Max Children = 15
Log Spam = yes (since we need the logs for stats.)
MS Spam Action is "attachment, deliver".
If one rbl goes down or is slow, we found that we very quickly had a
backlog of 20k messages, so we've turned off rbl checks in SA. Adjusting
the timeouts for rbl checks may be a better solution.

We've made a few other tweaks to the default MS settings, but I don't
think they are related to performance.

The load average when the machines are sharing the mail load is usually
about 3-4, rising to about 10-15 when one machine handles the full load
by itself. Everything seems to fit into memory, there is no swapping to
disk, and the average delay for a message being relayed is < 1 minute.

A possible bottleneck is syslog, since both sendmail & MS are very busy
adding log entries to the maillog, though I haven't done any testing to
see if this is really a problem.

Another bottleneck is the way 'user unknowns' are handled. Currently,
because mail is relayed through the MS boxes, the MS boxes don't know if
the user exists at the domain or not. So MS spends time scanning mail
that is destined for a user that doesn't exist. Next Sendmail tries to
deliver it and it is rejected by the destination mailserver. Then
sendmail tries to bounce it back to an address that usually doesn't
exist, or a mailserver that is not accepting connections or can't be
resolved. Using re-mqueue to re-queue outgoing mail is helpful with this
problem.

Ken A.


Tony Johansson wrote:
>>I am processing around 600.000 messages on two dual xeon machines, daily,
>>with peaks to 800.000-1.000.000 daily. I think MS is doing just fine :)
>>
>>Most of the time its a matter of the test setup also that is limiting the
>>figures. Also some tweaking on the machines wont harm...
>>
>>Bye,
>>Raymond.
>
>
> What MTA are you using? I recall an earlier post by Julian where he tested
> exim vs sendmail on one of his test machines, getting 3 times the
> throughput with exim.
>
> I'm interested in what kind of performance people are getting with
sendmail
> and MS. I'm about to design a system built on redhat and sendmail which
> will handle lots of relaying (500k-1mil email daily approx)
>
> I'd rather use sendmail but if the performance gain with exim really is 3
> to 1 that might be the road we haveto take.
>
> regards, Tony
>
>