Extremly slow scrub
Malte Schirmacher
i've seen my raidz reading and writing with more then 70MB/s.
But if i make it scrub the very same pool it's not reading more then
~2.5MB/s. Why?
Greetings
Malte
xyzzyx
Have you completed your scrub or have you stop it after a while of
slow scrubbing?
On my pool, when i start scrubbing, it says it will take more than 110
hours to finish, but it usually finish with 7 hours after starting.
It seems to pickup pace as it goes along.
This is my pool config
NAME SIZE ALLOC FREE CAP DEDUP HEALTH ALTROOT
ztank 6.31T 2.62T 3.69T 41% 1.01x ONLINE -
pool: ztank
state: ONLINE
scan: scrub repaired 21.5K in 6h53m with 0 errors on Sun Mar 6
17:51:40 2011
config:
NAME STATE
READ WRITE CKSUM
ztank
ONLINE 0 0 0
raidz1-0
ONLINE 0 0 0
cztank0
ONLINE 0 0 0
cztank1
ONLINE 0 0 0
cztank2
ONLINE 0 0 0
cztank3
ONLINE 0 0 0
cztank4
ONLINE 0 0 0
cztank5
ONLINE 0 0 0
cztank6
ONLINE 0 0 0
cache
scsi-SATA_OCZ-VERTEX2_OCZ-OOLXCWL74ZL36BLN-part1
UNAVAIL 0 0 0
cztankc ONLINE
10.3K 0 0
Malte Schirmacher
> Hi,
>
> Have you completed your scrub or have you stop it after a while of
> slow scrubbing?
> On my pool, when i start scrubbing, it says it will take more than 110
> hours to finish, but it usually finish with 7 hours after starting.
> It seems to pickup pace as it goes along.
Indeed i havn't scrubed completly yet, because i'm afraid of the kernel
panic which is likley to happen ;-)
But i should give it a try...
Malte Schirmacher
And as expected there is the kernel panic. It run a few hours (since the
last mail...).
Indeed it read with up to 18MB/s at the end. But now there is the panic
and it will be very hard to recover from it :-/
>
>
Clemens Fruhwirth
I bumped up my THREAD_ORDER in arch/x86/include/asm/page_64_types.h to
4, making the thread size 16k instead of the original 4k. I did that
in the hope that my bonnie++ panics will go away. To no avail.
However, zfs scrub on a small pool worked.
--
Fruhwirth Clemens http://clemens.endorphin.org
