svn commit: r300207 - in head: sbin/camcontrol sys/cam/ata sys/cam/scsi sys/dev/ahci sys/geom sys/geom/eli sys/kern sys/sys usr.sbin usr.sbin/diskinfo usr.sbin/zonectl

[Alan Somers]

On Thu, May 19, 2016 at 9:54 AM, Adrian Chadd <adrian.chadd at gmail.com>
wrote:

> On 19 May 2016 at 07:50, Kenneth D. Merry <ken at freebsd.org> wrote:
> > On Thu, May 19, 2016 at 15:30:37 +0100, Bruce Simpson wrote:
> >> Right On Commander!
> >>
> >> Now here is hoping the random-write issues with OpenZFS are solved...
> >
> > Well, this is of course only the underlying infrastructure for SMR
> drives.
> >
> > As for the random write issues with ZFS (or rather sequential I/O going
> out
> > in random order), we (Spectra) are working on that in general.
> >
> > The original plan was to do full SMR support for ZFS, but it looks like
> we
> > may get good enough write performance with Drive Managed drives by just
> > making writes from ZFS to drives more sequential rather than going for
> full
> > Host Aware and Host Managed support.  If that turns out that performance
> > isn't good enough after all, then we'll have to put in the effort to put
> > Host Aware support at least into ZFS.
> >
> > That's why I said (in the commit message) that I don't know of anyone
> > working on using the SMR infrastructure for UFS or ZFS.  Now that the
> > plumbing is in place in CAM and GEOM, it is possible at the filesystem
> > level if someone wants to do the work.
>
> Is it possible to use this from userland for now? Ie, probe whatever
> we need to in order to understand the transaction geometries (ie, the
> optimal sizes for reading/writing, etc) and then issue IOs that are in
> line with this?
>
>
>
> -adrian
>
>
Sure.  You can do all that with zonectl or its ioctls, if you have a
userland program that writes to raw block devices.  Reset Write Pointer is
supported, too.

-Alan