[OmniOS-discuss] [discuss] background snapshot destruction impact

[Richard Elling]

On Dec 22, 2013, at 4:23 PM, Tobias Oetiker <tobi at oetiker.ch> wrote:

> Hi Richard,
> 
> Yesterday Richard Elling wrote:
> 
>>> c) shouldn't the smarter write throttle change
>>>  https://github.com/illumos/illumos-gate/commit/69962b5647e4a8b9b14998733b765925381b727e
>>>  have helped with this by makeing zfs do its internal things
>>>  with a lower priority.
>> 
>> Yes, but the default zfs_vdev_max_pending remains at 10. Once
>> the I/Os are sent to disk, there is no priority scheduling. You
>> should consider lowering zfs_vdev_max_pending to allow the ZIO
>> scheduler to do a better job of rescheduling the more important
>> I/Os.
> 
> the patch mentioned introduces a ton of new tuneables but it
> removes zfs_vdev_max_pending

Indeed, these are now zfs_vdev_max_active and friends.
It is very unclear to me what your graphite is attempting to show.
Is this data from the pool itself, or from vdevs under the pool?
The pool-level stats are mostly useless for this analysis, need to
see the per-vdev stats.
 — richard