[zfs-discuss] Re: 6x drive raidz1 Ubuntu 12.04 performance problems

Gordan Bobic gordan.bobic at gmail.com
Thu Jun 7 16:29:58 EDT 2012


On 07/06/2012 18:56, DC wrote:
> ~$ iostat -x 2
> Linux 3.2.0-24-generic (bondi)  06/07/2012      _x86_64_        (2
> CPU)
>
> avg-cpu:  %user   %nice %system %iowait  %steal   %idle
>             2.59    1.99    7.40    2.70    0.00   85.32
>
> Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s
> avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
> sda               1.28     0.02  114.54   12.70  3878.27    24.14
> 61.34     0.62    4.87    4.66    6.81   1.19  15.18
> sdb               1.11     0.02  115.50   13.20  3881.01    25.57
> 60.71     0.66    5.12    5.09    5.36   1.20  15.44
> sdc               1.28     0.04  114.54   13.01  3882.14    25.42
> 61.27     0.70    5.50    5.26    7.66   1.28  16.36
> sdd               0.00     0.17    0.01   80.36     0.05  3758.27
> 93.53     4.94   61.51    2.12   61.51  10.49  84.31
> sde               1.23     0.02  113.01   12.72  3883.60    25.45
> 62.18     0.54    4.32    4.47    3.04   1.04  13.08
> sdf               1.23     0.02  112.44   12.18  3878.86    24.17
> 62.64     0.51    4.06    4.15    3.20   1.00  12.45
> sdj               1.98     0.04  102.54   12.73  3884.18    24.11
> 67.81     0.73    6.33    6.46    5.29   1.59  18.36
>
> sdd is the new replacement drive, sdj is the problematic one.  I must
> say, in general, things look consistent.  sdj hasn't looks poor since
> the resilver started

Again, these figures seem OK - you are getting ~110 IOPS/disk which is 
spot on where I'd expect it to be. I think you'll have to live with it - 
your data just got too fragmented so resilvering is making the disks 
seek all over the place.

Perhaps there is an optimization to be made here in ZoL WRT resilvering 
approach. Is there scope for re-arranging the order of resilvering so it 
is resilvering stripes in on-disk order?

Gordan



More information about the zfs-discuss mailing list