[Dovecot] very slow imap opens in pine after moving maildirs to new drive
Hi,
I've been running dovecot-0.99.10 for the past several years, with virtually no problems. Lately, I noticed that the filesystem holding my users' imap maildirs was getting close to full. So... last night, I used tar to copy /var/spool/imap to a new drive (9G scsi, reiserfs) that I then mounted under /var/spool/imap (after moving the old dir aside).
After the move, what would take pine a handful of seconds, with the load average barely breaking .40, now takes several minutes, with a load average spike to nearly 3.
I must admit to being rather puzzled by this, as I would not have expected the move to result in worse performance. Do I need to figure out what the index file names are, and delete those? Any suggestions greatly appreciated - as even closing pine now takes nearly 5 minutes, whereas when dovecot talks to my old imap dir (I've tested this), closing the session takes no more than 10 seconds.
server is Mandrake 9.1, 2.4.21-0.13mdkenterprise, P4 2.6GHz, 1G RAM 9G U2W SCSI system disk, 9G U2W SCSI (and big*ss IDE drive for /users). Each of the scsi drives is on its own aha-2940UW controller.
Filesystem on old imap location (/var filesystem) is reiserfs, same as that on new dedicated disk.
Thanks much,
Dave
--
Dave Lugo dlugo@etherboy.com LC Unit #260 TINLC Have you hugged your firewall today? No spam, thanks.
Are you the police? . . . . No ma'am, we're sysadmins.
Dave Lugo wrote:
Hi,
I've been running dovecot-0.99.10 for the past several years, with virtually no problems. Lately, I noticed that the filesystem holding my users' imap maildirs was getting close to full. So... last night, I used tar to copy /var/spool/imap to a new drive (9G scsi, reiserfs) that I then mounted under /var/spool/imap (after moving the old dir aside).
After the move, what would take pine a handful of seconds, with the load average barely breaking .40, now takes several minutes, with a load average spike to nearly 3.
I must admit to being rather puzzled by this, as I would not have expected the move to result in worse performance. Do I need to figure out what the index file names are, and delete those? Any suggestions greatly appreciated - as even closing pine now takes nearly 5 minutes, whereas when dovecot talks to my old imap dir (I've tested this), closing the session takes no more than 10 seconds.
server is Mandrake 9.1, 2.4.21-0.13mdkenterprise, P4 2.6GHz, 1G RAM 9G U2W SCSI system disk, 9G U2W SCSI (and big*ss IDE drive for /users). Each of the scsi drives is on its own aha-2940UW controller.
Filesystem on old imap location (/var filesystem) is reiserfs, same as that on new dedicated disk.
tried:
. rm'ing .imap.index* no improvement noted
. putting both scsi disks on same controller no improvement noted
I'm about to try placing the maildirs on the IDE disk used for /users, to see if that makes a difference.
FWIW, Thunderbird (even after I remove the .msf file for my inbox) still opens MUCH faster than pine; less than a minute for the former, vs nearly 5 minutes for the latter.
I'm getting rather more concerned, because if dovecot refuses to play nice on all but the fs I originally saved the maildirs to, I'm going to run out of space within days.
--
Dave Lugo dlugo@etherboy.com LC Unit #260 TINLC Have you hugged your firewall today? No spam, thanks.
Are you the police? . . . . No ma'am, we're sysadmins.
On Mon, 2 Jan 2006, Dave Lugo wrote:
I'm about to try placing the maildirs on the IDE disk used for /users, to see if that makes a difference.
That worked. I can only assume something odd about that second scsi disk is the issue, but there are no unusual log messages, nor does a 'verify disk' via the controller's bios turn up any bad sectors.
My apologies for the list noise. Initially, this sure looked like a dovecot issue...
--
Dave Lugo dlugo@etherboy.com LC Unit #260 TINLC Have you hugged your firewall today? No spam, thanks.
Are you the police? . . . . No ma'am, we're sysadmins.
Am Montag, 2. Januar 2006 21:15, schrieb Dave Lugo:
That worked. I can only assume something odd about that second scsi disk is the issue, but there are no unusual log messages, nor does
You could try a benchmark to check this disk's performance and CPU utilization needed to transfer data to it, eg. "bonnie" or something similar...
Greetings,
Gunter
-- +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
- http://aachen.uni-dsl.de/ - Der direkte Draht in's Hochschulnetz! + +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ "The thing is that Mr. Dibbler can even sell sausages to people who have bought them off him *before*." -- (Terry Pratchett, Moving Pictures) +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+PGP-verschlüsselte Mails bevorzugt! +
Dave Lugo wrote:
users' imap maildirs was getting close to full. So... last night, I used tar to copy /var/spool/imap to a new drive (9G scsi, reiserfs) that I then mounted under /var/spool/imap (after moving the old dir aside).
What was the old fs? I remember something in the reiserfs FAQ about bad performance if you tar-copied from some other fs... ah, yes: http://www.namesys.com/faq.html#file-ordering-speed
participants (3)
-
Dave Lugo
-
Gunter Ohrner
-
Jakob Hirsch