btrfs + luks cause serious CPU usage

drahnr · 2011-06-29 11:00:34

I often tried to forget about this, but in a time I use a external disk with btrfs.

bus as soon as I mount the disk via luks mapper device, I get serious CPU usage by kthreads. It goes that far that even typing into pidgin gets lagged/delayed.

Does anybody else have this kind of issue? An internal disk has the verysame effect. Tested copies with these combinations: ext4→btrfs; btrfs→btrfs; btrfs→ext4

Thank you!

Appendix:

Phenom 9500 4x 2.2
8GB DDR2 SDRAM
...
so the system should be pretty much capable of anything :>

Last edited by drahnr (2011-06-29 11:01:43)

mikesd · 2011-06-29 11:47:50

I'm using an external USB2 disk as a backup device and use btrfs on luks. The disk is mounted all the time and I don't see excess CPU usage as a result of this.

I just tried using dd to write to the btrfs fs and it did peg one of my two cores at 100% which continued for several seconds after I killed the job.

Is it possible that something is accessing the disk such as a file indexer?

drahnr · 2011-06-29 14:14:49

I do not use any indexer/tracker, I do hourly backups but that should not be the reason for ongoing CPU utilization near max as these are delta backups and occur only once an hour.

Any ideas?
What mount options do you use for your btrfs partition?

Excerpt from /etc/fstab

/dev/mapper/luks_foo        /foo        btrfs defaults 0 1

Last edited by drahnr (2011-06-29 14:18:34)

mikesd · 2011-06-30 01:14:51

I use pretty much the same options as you:

$ grep backup /etc/fstab
/dev/mapper/backup /mnt/backup btrfs defaults,noauto 0 0

I use the standard kernel from testing which is 2.6.39 though I have not seen this behavior on any previous kernel either.

If you don't use an indexer and something like iotop shows no activity I'm not sure what to try next. If you have hourly backups in place how many snapshots are there present on the volume? Are they being rotated or just building up? Could btrfs be doing some sort of rebalance (seems unlikely on a single disk pool)? Probably not being much help sorry.

drahnr · 2011-06-30 11:18:30

The backup disk is ext4 formatted and it rotates/renames the snapshots. It shoulnd't be to ebil, rnsapsot is essentially rsync with some bash fun. I use the stable 2.6.39 kernel, and I experienced this with a lot of kernels before too. I will give iotop a shot, did not know about that. Thank you! (though not fixed yet)

Last edited by drahnr (2011-06-30 11:39:31)

drahnr · 2011-11-14 23:49:03

I did a fresh setup (for another reason) and this obviously fixed it so far. The only change: i do not use an encrypted root disk anymore. I will check if an encrypted root was the cause be resetting it again (well actually just moving the root content to another luksdisk) so i can do an inplace comparsion. I'll report back as soon as I am done.

Last edited by drahnr (2011-11-15 00:00:26)

Arch Linux

#1 2011-06-29 11:00:34

btrfs + luks cause serious CPU usage

#2 2011-06-29 11:47:50

Re: btrfs + luks cause serious CPU usage

#3 2011-06-29 14:14:49

Re: btrfs + luks cause serious CPU usage

#4 2011-06-30 01:14:51

Re: btrfs + luks cause serious CPU usage

#5 2011-06-30 11:18:30

Re: btrfs + luks cause serious CPU usage

#6 2011-11-14 23:49:03

Re: btrfs + luks cause serious CPU usage

Board footer