bill.martin at ssi.samsung.com
Wed Apr 15 09:23:51 PDT 2015
* From the T10 Reflector (t10 at t10.org), posted by:
* Bill Martin-SSI <bill.martin at ssi.samsung.com>
It is important for the specifics addressed by the stream identifier to know
when the stream identifier becomes valid and when it is no longer valid in
order to be able to re-use the stream identifier and know when keeping track
of the resources pointed to by the stream identifier will not have any future
writes to that group of resources.
I think that you may be more interested in the per IO hints that Fred is
SSD I/O Standards
Samsung Semiconductor, Inc.
Cell (408) 499-1839
From: Martin K. Petersen [mailto:martin.petersen at oracle.com]
Sent: Monday, April 13, 2015 7:33 PM
To: Kevin D Butt
Cc: Bill Martin-SSI; T10 org
Subject: Re: T10/14-275r8 question
>>>>> "Kevin" == Kevin D Butt <kdbutt at us.ibm.com> writes:
Kevin> Why does the storage intelligence restrict the Background
Kevin> Operation Control and Write Stream to only Resource Provisioned
Kevin> devices, when there could be value if it was used for both Thin
Kevin> and Thick Provisioning?
I agree that the stream concept is generally useful for I/O classification
and should not be tied to any particular type of device provisioning. We have
been wanting this for a long time.
With the OS hat on I would like to share as much information as I possibly
can about the nature of the I/O request I am sending to the target. The
device is then free to use that information in whichever way may cause it to
optimally manage its internal resources. Regardless of whether it is a disk
drive, array or solid state device. And if the device can't make use of the
information I provided then it is free to ignore it.
As a result, I am not a fan of the extra effort of having to explicitly open
and close streams. I would much prefer to see a fixed maximum based on CDB
field size and the storage device is then free to interpret passed-in stream
identifier as it pleases (akin to GROUP NUMBER but large enough to be
useful). A bigger WRITE (and READ) CDB to accommodate the additional
information is fine with me.
Martin K. Petersen Oracle Linux Engineering
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org
More information about the T10