I introduced a bug in 15-113

Ralph Weber Ralph.Weber at wdc.com
Mon Jun 8 08:12:10 PDT 2015


Formatted message: <a href="http://www.t10.org/cgi-bin/ac.pl?t=r&f=r1506081_f.htm">HTML-formatted message</a>

Joe,
Now that I've got you sipping the Kool Aid, how about a big gulp?
Let:
  *   MAX = MAXIMUM NUMBER OF EXPLICITLY OPEN SEQUENTIAL WRITE PREFERRED
ZONES
  *   OPT = OPTIMAL NUMBER OF OPEN SEQUENTIAL WRITE PREFERRED ZONES
  *   IPOP = current number of implicitly opened zones (write preferred, of
course)
  *   EPOP = current number of explicitly opened zones
How about scaling the max against the number of implicit opens? Basically,
the only way the host gets the full max is to have no implicitly opened
zones.
Assuming that I can still to algebra, the truth test (a al the current ZBC)
looks something like:
error OPEN ZONES command if ... EPOP > MAX * (1 - ( IPOP / OPT ))
All the best,
.Ralph
________________________________
From: Joe Breher [Joe.Breher at hgst.com]
Sent: Monday, June 08, 2015 9:55 AM
To: Ralph Weber
Cc: T10 org
Subject: Re: I introduced a bug in 15-113
Thanks Ralph -
Indeed, my earlier mention of a MAXIMUM NUMBER OF EXPLICITLY OPEN SEQUENTIAL
WRITE PREFERRED ZONES was meant to imply such a solution. I was hopeful that
there may be some other solution that did not require adding a new VPD
parameter at this late date. However, no better solution seems immediately
apparent.
Joe Breher
Storage Architecture Technologist
Standards Setting Organization
San Jose Research Center
HGST, a Western Digital company
(478) 2-Breher
(478) 227-3437
This e-mail may contain confidential or legally privileged information of
HGST. If you are not the intended recipient, please notify us immediately by
responding to this e-mail and then deleting it from your system.
On Jun 6, 2015, at 5:47 AM, Ralph Weber
<Ralph.Weber at wdc.com> wrote:
Joe,
Caveat Emptor: I have no skin in the Sequential Writes Preferred game. The
following suggestion is completely motivated by the notion that Sequential
Writes Preferred is a way station between Conventional zones and Sequential
Writes Required zones. It may not (or may) satisfy the actual needs on the
ground.
Perhaps the thing to do in the Sequential Writes Preferred case is to
hard-limit the number of explicitly opened zones ... with the implicitly
opened zones continuing to have the "optimal" number of open zones behavior.
This would constrain the error returns to OPEN ZONE commands and thus come
closer to preserving the desired backwards compatibility that Sequential
Writes Preferred is intended to provide.
Remember: This is only a suggestion for consideration by the experts in the
field.
All the best,
.Ralph
________________________________
From: owner-t10 at t10.org<mailto:owner-t10 at t10.org>
[owner-t10 at t10.org<mailto:owner-t10 at t10.org>] on behalf of Joe Breher
[Joe.Breher at hgst.com]
Sent: Friday, June 05, 2015 5:39 PM
To: Gerry Houlder
Cc: T10 org
Subject: Re: I introduced a bug in 15-113
Thanks for your thoughts, Gerry.
However, I find no support for your claim that your option (b) is 'the
thought that is currently in the standard'. As countervailing examples,
 3.1.30 open zone resources
zone resources that are used only by a zone with the zone condition set to
IMPLICIT OPEN or set to EXPLICIT
OPEN
and
 4.3.3.2.4.1 ZC3:Explicit_Open state overview
While in this state:
a) the device server allocates open zone resources:
A) to maintain the accuracy of the write pointer; and
B) for the availability of media to support write operations to this zone;
>From these, we see that zones in the ZC3:Explicit_Open under no
circumstances have their open zone resources stolen out from underneath them.
>From this requirement, my interpretation is that your option (b) is not only
not 'the thought that is currently in the standard', but is also explicitly
prohibited.
Further, sequential write preferred is not documented as accepting and
processing everything the host sends. Specifically, the zone management
operation selects a zone to be closed only under a certain set of criteria
(actually 'may select'). While it is unspecified what will happen when that
criteria is violated (which is the hole I am trying to close), there is also
no 'shall' requirement of which I am aware that mandates such successful
completion. If I have merely missed it, I would be happy to be shown the
error of my ways. Well, I probably wouldn't be 'happy', but I'd begrudgingly
accept it.
Sure, there is the guiding principle that host aware devices should 'just
work' in legacy systems. However, this principle is inapplicable to the
problem at hand, due to the fact that such legacy systems will not ever send
a OPEN ZONE command.
I still think we have technical work to do here. Still hoping for a less
intrusive change than that sketched out earlier.
Joe Breher
Storage Architecture Technologist
Standards Setting Organization
San Jose Research Center
HGST, a Western Digital company
(478) 2-Breher
(478) 227-3437
This e-mail may contain confidential or legally privileged information of
HGST. If you are not the intended recipient, please notify us immediately by
responding to this e-mail and then deleting it from your system.
On Jun 5, 2015, at 3:14 PM, Gerry Houlder
<gerry.houlder at seagate.com> wrote:
There are two ways to think about this for sequential write preferred.
(a) Explictly opened zones have a maximum limit because there is a desire to
promise no performance loss for zones that are explicitly opened. This limit
is probably higher than the optimal number of open zones or it might be the
same. If the disk drive has to prevent new explicit opens from happening,
then OPEN ZONE commands are rejected just like they would be on sequential
write required zones.
(b) The other option (which is the thought currently in the standard) is that
sequential write preferred provide no promise of better performance for
particular zones once the optimal number is exceeded. if the host insists on
opening more zones (either implictly or explicitly) without actions to close
or fill some zones, then the disk drive allocates performance resources to
whatever zone(s) seem to be most active in a vendor specific manner
(regardless of their implicit or explict attribute).
Behavior (a) introduces a "more likely to want better performance" hint that
is absent from the ZBC standard. However, sequential write preferred is
already documented to accept and process everything the host sends and if the
host willfully ignores the drive's reported optimal limits then some commands
will have lower performance than would usually be expected and the drive
becomes the arbiter of how to maintain best performance. Going above the
optimal limit should cause the drive to react as if all open zones are
implicitly open, but none will be implicitly closed.
On Fri, Jun 5, 2015 at 3:07 PM, Joe Breher
<Joe.Breher at hgst.com> wrote:
Hi Ralph -
I hate to be disagreeable, but I disagree with your disagreement.
The page 9 comment I wish to retract is not in 4.3.3.op.6.3, but in
4.3.3.op.6.2  Select a sequential write preferred zone, where I commented
"All zone management operations to sequential write preferred zones return
successful completion."
There is a very real limit on the number of concurrent explicitly open
sequential write preferred zones a device is able to support - for at least
some implementations. If this limit is reached, and there are no implicitly
open sequential write preferred zones, the device must fail the zone
management operation, as the guiding directive is that zone management
operations shall not cause explicitly open zones to be closed. Note the (y >
0) term in the expression.
I realize (now) that we currently have no mechanism to report such a limit.
This may be the root issue with this feature. We are still confabbing
internally with the proper resolution of this -- which is why I have not yet
forwarded proposed text -- but at least one solution is immediately apparent.
Unfortunately, it introduces a new technical requirement at a late date. This
would be to introduce a new VPD parameter of MAXIMUM NUMBER OF EXPLICITLY
OPEN SEQUENTIAL WRITE PREFERRED ZONES, and have a secondary test in the
'Select a sequential write preferred zone' path.
Suggestions for a less intrusive resolution are welcome.
Joe Breher
Storage Architecture Technologist
Standards Setting Organization
San Jose Research Center
HGST, a Western Digital company
(478) 2-Breher
(478) 227-3437<tel:%28478%29%20227-3437>
This e-mail may contain confidential or legally privileged information of
HGST. If you are not the intended recipient, please notify us immediately by
responding to this e-mail and then deleting it from your system.
On Jun 4, 2015, at 6:59 PM, Ralph Weber
<Ralph.Weber at wdc.com> wrote:
Joe,
I disagree with your retraction of the page 9 comment.
The only way an error is returned is if the second equation in 4.3.3.op.6.3
Select a sequential write required zone evaluates to true. If the values in
the equation are spelled out, the following must evaluate to true:
the number of sequential write required zones with a Zone Condition of
EXPLICITLY OPENED
ge
the contents of the MAXIMUM NUMBER OF OPEN SEQUENTIAL WRITE REQUIRED ZONES
field of the Zoned Block Device Characteristics VPD page
Now, if there is so much as one sequential write preferred zone open, then
the zoned device is not allowed to support any sequential write required
zones. Therefore, the number of sequential write required zoned with a Zone
Condition of EXPLICITLY OPENED must be zero.
Surely, zero should be a small enough number to be less than any valid
contents of the MAXIMUM NUMBER OF OPEN SEQUENTIAL WRITE REQUIRED ZONES field,
n'es pas?
All the best,
.Ralph
________________________________
From: Joe Breher [Joe.Breher at hgst.com]
Sent: Wednesday, June 03, 2015 10:45 AM
To: Ralph Weber; T10 org
Cc: Gerry Houlder
Subject: ZBC: I introduced a bug in 15-113
I just realized that I introduced a bug in 15-113r4 as modified.
On page 9, I commented "All zone management operations to sequential write
preferred zones return successful completion".However, in the case where
there are no implicitly opened zones available to be closed (i.e., all open
zones are explicitly opened zones), the zone management operation will indeed
return a non-successful completion.
I will have suggested text to correct this later today.
Joe Breher
Storage Architecture Technologist
Standards Setting Organization
San Jose Research Center
HGST, a Western Digital company
(478) 2-Breher
(478) 227-3437<tel:%28478%29%20227-3437>
This e-mail may contain confidential or legally privileged information of
HGST. If you are not the intended recipient, please notify us immediately by
responding to this e-mail and then deleting it from your system.



More information about the T10 mailing list