floored with the performance after migration

devsk devsku at gmail.com
Sun Apr 24 17:47:52 EDT 2011


The system is idle as far as this pool is concerned. Its not even
rootfs in this case. The writes are some sort of ZFS updates while
scrubbing the data. I frankly speaking don't know.

-devsk


On Apr 24, 2:34 pm, Brian Behlendorf <behlendo... at llnl.gov> wrote:
> That may be a good explanation, particularly for the scrub performance.
> If there is other ZFS I/O the scrub will back off giving priority to the
> user I/O.  Of course the spurious I/O your seeing needs to be explained,
> I haven't seen this behavior yet on my test systems...  clearly you have
> a more interesting situation.
>
> --
> Thanks,
> Brian
>
>
>
>
>
>
>
> On Sat, 2011-04-23 at 23:44 -0700, devsk wrote:
> > Looks likehttps://github.com/behlendorf/zfs/issues/210have something
> > to do with it. What I notice in zpool iostat output is that whenever
> > there is a write to the RAID, the read speed falls down considerably
> > and then picks up again.
>
> > -devsk
>
> > On Apr 23, 10:44 pm, devsk <dev... at gmail.com> wrote:
> > > I just migrated my data pool from zfs-fuse to native ZFS. And I am
> > > floored by the performance. The scrub is going at 1/3rd to half of the
> > > speed achieved by zfs-fuse. More tests and comparisons will follow.
>
> > > I can't explain this at this time. Any ideas?
>
> > > -devsk



More information about the zfs-discuss mailing list