On 11-Feb-01 at 11:39, Andy Hubbell, Jr. ([EMAIL PROTECTED]) wrote:
> I'm not seeing any evidence that postfix is batching up the local
> deliveries through a single lmtp instance... I've made some changed
> recommended by others using postfix, but still don't see evidence of this
> happening... Not sure if its my postfix config or my cyrus config that's
> causing this...
The Cyrus config shouldn't be able to affect batching through the
LMTP protocol. I know this works with Exim - I just finished
testing a Mailman/Exim/Cyrus setup and I've examined some typical
protocol logs for the LMTP dialog. (A couple of your other recent
messages to the list make me think you are on the right track with
postfix config changes to make this work.)
In a previous message you asked about the Cyrus master starting a
new lmtpd for each delivery. Check into the prefork clause of the
cyrus.conf file - tweaking that might help.
> I've been watching things closely for a couple weeks here, and I'm sure
> that amavis/AVP (the virus scanner) is using the majority of the
> resources... Unfortunately virus scanning every email in and out is
> somewhat expensive. I've modified it so as to use the
> AVPDaemon/AVPDaemonClient process which improved things tremendously. But
> given the proliferation of viruses in our user community, I don't dare turn
> off that! This thing catches several hundred virii each day!
Anything you can do to speed that up would help tremendously. It looks
like you have verified that it only happens once for each incomming
message; which is the most important factor. You might also see if there's
any way to skip it entirely if the message doesn't include any attachments.
(Or HTML if amavis looks for VBScript/JavaScript viri.)
> I do have duplicate delivery suppression turned on, and I have the Berkeley
> DB3 package installed so I thought it was using that... Not certain how
> to verify it... Hmm, checked with 'file' and my deliver-?.db's seem to be
> "Berkeley DB 2.X Hash/Little Endian"....
I think I saw something on one of the mailing lists about reports of
some performance problems with db3 under some circumstances; but I
don't know anything at all about the details.
> Last night I converted everything over to a fully hashed directory
> structure, because before everything seemed to be hashed except for the
> default partition where the mail is actually stored. That should take care
> of linux's issues with large directories.
That's a good start for a large installation like yours. There may
also be some filesystem attributes that you could tweak to improve
performance for a busy filesystem. Check out the options that are
suggested for usenet partitions; but be a little careful about adopting
all of them - some of them can cause loss of data if the system goes
down suddenly. That's acceptable for netnews; but not for mailboxes.
-Pat
