[lug] imap performance
nate at natetech.com
Fri Feb 7 14:59:03 MST 2003
How much RAM is in the mail server?
I found that even on a lightly loaded UW-IMAP machine, adding physical RAM
made a world of difference in the responsiveness of the server. I never dug
into the code to see if it's pulling a lot of things into RAM when there are
a lot of mail files, and this particular box doesn't see many simultaneous
users, but all the users have a lot of mail sorted and stored on it in
Also, are you running imap through inetd/xinetd or as a direct server daemon
listening on a port? inetd/xinetd could slow down the process considerably.
On this lightly loaded server, I decided I still wanted some of the features
of xinetd in front of UW, so I left it there, but I did find that certain
clients (Outlook Express in particular) if a user had a LOT of mail folders
could make more connections per minute than inetd would allow to the IMAP
daemon... and had to raise the limits in the configuration.
The symptom was people getting "the server has unexpectedly disconnected
you" messages, when really inetd was killing off imap... (GRIN). Heh...
only took about ten seconds of looking at the logs to find that one...
Nate Duehr, nate at natetech.com
----- Original Message -----
From: "Hugh Brown" <hugh at math.byu.edu>
To: "LUG" <lug at lug.boulder.co.us>
Sent: Friday, February 07, 2003 2:29 PM
Subject: [lug] imap performance
> I have users complaining about slow performance of the mail server. I'm
> using UW-IMAPD. Usually they have large mail spools (large => greater
> than 50MB and about 4000-5000 messages). Should I tell them to file
> their mail and pare down their inboxes, or should I look for something
> Alternatively, is there an optimal size for mailbox files? Is there a
> size that doesn't cause massive slow downs, but doesn't require having
> lots of little mailbox files
> Web Page: http://lug.boulder.co.us
> Mailing List: http://lists.lug.boulder.co.us/mailman/listinfo/lug
> Join us on IRC: lug.boulder.co.us port=6667 channel=#colug
More information about the LUG