I'm investigating a performance issue and trying to determine if something
might be funny with ZFS. I have two pools, each are 6x300G SCSI in raidz1
config. Zpool version 15. Each pool has a single filesystem (so 2 total).
'zpool iostat -v 5' shows the following pretty consistently:
capacity operations bandwidth
pool used avail read write read write
----------- ----- ----- ----- ----- ----- -----
fast 1.11T 529G 113 0 472K 0
raidz1 1.11T 529G 113 0 472K 0
c4t0d0 - - 83 0 4.10M 0
c4t1d0 - - 80 0 4.21M 0
c4t2d0 - - 86 0 4.40M 0
c4t3d0 - - 81 0 4.27M 0
c4t4d0 - - 81 0 4.06M 0
c4t5d0 - - 78 0 4.22M 0
----------- ----- ----- ----- ----- ----- -----
fast2 1.05T 586G 105 0 509K 0
raidz1 1.05T 586G 105 0 509K 0
c4t8d0 - - 81 0 3.90M 0
c4t9d0 - - 77 0 3.76M 0
c4t10d0 - - 83 0 3.90M 0
c4t11d0 - - 78 0 3.79M 0
c4t12d0 - - 79 0 3.72M 0
c4t13d0 - - 77 0 3.76M 0
----------- ----- ----- ----- ----- ----- -----
Why would the per-disk bandwidth be so much higher than the bandwidth for the
top-level device? Should this be expected? How can I investigate further?
I have atime disabled on both pools, zpool status shows no errors, and prtpicl
shows the sync speed for all disks to be 320k (as expected for SCSI).
CPU load is very low, network link appears OK.
The report from my user is that content is being processed at 80dps (docs/sec)
from this system, where a different box with all 12 disks in a single raidz1 is
doing 200dps. Additionally, the faster system only has 4GB memory whereas the
slower one has 16G. (I know, "remove the extra memory!")
Any help is appreciated - I only know enough Solaris to be dangerous. :)
Thanks,
Dan
_______________________________________________
Discuss mailing list
[email protected]
http://lists.blu.org/mailman/listinfo/discuss
