-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Hi,
I've got an 8 disk raid 6 array, and I've noticed recovery after adding a new
disk with 7/8 disks active seems to be sub-optimal (with 6/8 disks active it
seems fine). To be more specific, with 7/8 disks active it seems to do extra
reads from one of the active drives, slowing the recovery down to about
30MB/sec on my machine (with 6/8 disks it recovers at about 50MB/sec). Here
is the output of "iostat -mx 10 /dev/sd?" during a recovery with 7/8 disks
active (see the line for sdf in particular):
avg-cpu: %user %nice %system %iowait %steal %idle
0.00 0.00 13.48 0.00 0.00 86.52
Device: rrqm/s wrqm/s r/s w/s rMB/s wMB/s avgrq-sz
avgqu-sz await svctm %util
sda 0.00 6365.60 0.00 1292.70 0.00 29.90 47.38
1.61 1.25 0.38 48.52
sdb 6876.40 0.00 784.70 0.00 29.90 0.00 78.02
2.14 2.73 0.52 40.76
sdc 6868.70 0.00 793.30 0.00 29.94 0.00 77.29
1.92 2.44 0.47 37.60
sdd 6850.20 0.00 811.30 0.00 29.91 0.00 75.50
2.09 2.58 0.48 38.84
sde 6860.80 0.00 800.20 0.00 29.90 0.00 76.52
2.39 2.99 0.54 42.88
sdf 7953.50 0.00 664.10 0.00 33.60 0.00 103.63
12.46 18.68 1.51 100.00
sdg 6851.20 0.00 810.50 0.00 29.93 0.00 75.62
1.91 2.37 0.46 37.16
sdh 6836.40 0.00 825.20 0.00 29.91 0.00 74.24
2.36 2.87 0.50 41.04
Most of the drives are being read at 30MB/sec, and the new drive is being
written at the same speed. However, sdf is being read at 34MB/sec, which
puts its utilization at 100% (I guess due to the extra seeks) and is the
limiting factor in the recovery (the other drives are at less than 50%
utilization).
In contrast, when only 6/8 drives are active, all the active drives are read
from evenly at 50MB/sec, and the new drive is written at the same rate.
Is this a known issue? From what I understand of raid 5/6, I don't know of
any reason it would need to do extra reads from one drive.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
iEYEARECAAYFAkrAuqQACgkQ5vihyNWuA4XGiACaAhww4HQLQp3a/73s1WhHMpxL
0PwAoOPZUVn8X5FFn/bmsBNEk9/6HaaU
=zlB7
-----END PGP SIGNATURE-----