Question about UNMAP command

James C Hatfield james.c.hatfield at seagate.com
Wed Jan 28 09:49:05 PST 2015


Formatted message: <a href="http://www.t10.org/cgi-bin/ac.pl?t=r&f=r1501281_f.htm">HTML-formatted message</a>

Gerry,
ATA devices are assumed to be fully provisioned, and have no restriction on
support for the DATA SET MANAGEMENT 'Trim' command.
So, by extension, fully provisioned SCSI devices should be able to support
UNMAP as well.	(Yes: I know "SCSI rules don't apply to ATA", but a
precedent has been set.)
Thank You !!!
-----------------------------------------------------------------
Jim Hatfield
Seagate Technology LLC
   e-mail:  James.C.Hatfield at seagate.com
   s-mail:  389 Disc Drive;  Longmont, CO 80503 USA
   voice:  720-684-2120
   fax....: 720-684-2711
   cell...: 720-771-8914
On Wed, Jan 28, 2015 at 8:45 AM, Gerry Houlder <gerry.houlder at seagate.com>
wrote:
> Some questions have come up about whether a fully provisioned logical unit
> may support the UNMAP command. SBC-4 r4 has a table (table 5) in the
> Logical Block Provisioning Model that states that fully provisioned logical
> units are prohibited from supporting anchored or deallocated states. The
> wording in 4.7.2 says a fully provisioned logical unit may support GET LBA
> STATUS command but says nothing about whether UNMAP may be supported. This
> suggests that a fully provisioned logical unit may support UNMAP as long as
> no LBAs are anchored or deallocated. Does the group agree with this
> interpretation?
>
> A reason to do this is that more host systems are using UNMAP to indicate
> files they are deleting. These systems often send the UNMAP based solely on
> the fact that the file is being deleted, not on the type of logical unit
> that is in use. These systems would rather have a fully provisioned logical
> unit act like a provisioned logical unit (i.e., accept the UNMAP command,
> do nothing, and return GOOD status) than to reject the command as
> unsupported. This should not be prohibited by the standard.
>
> Another situation where this can come up is with a Host Aware ZBC device.
> These devices are required to return a default initialization pattern for
> any unwritten LBAs. To a host system, this looks just like those LBAs have
> been unmapped by the device. Does this mean that a Host Aware device has to
> call itself a resource provisioned device or can it still be a fully
> provisioned device even though it appears to implement anchoring? What if
> that device also accepts UNMAP commands but doesn't do anything except
> return GOOD status on that command?
>
>
>



More information about the T10 mailing list