Re: [storm] Plan for iSCSI work

[Mallikarjun Chadalapaka <[e-mail]>]

Hi Hemal,

>1.	The implementation of TaskReporting is required and NotUnderstood
response for TaskReporting key is not permitted

No.  Note that TaskReporting is not defined in RFC 3720.  Patrick MacArthur
proposed on this list a couple of weeks ago that we should fix this
disconnect in the consolidated draft - which I agreed with because I think
it's a good suggestion.  

 >2.	To enable FastAbort on a session, the negotiation of TaskReporting
key is required.

Correct, specifically "TaskReporting=FastAbort" must be negotiated.

>3.	Only when the TaskReporting=FastAbort functionality is supported,
the protocol behavior specified in Section 4.2.3.4 is required to be
implemented and exhibited.

If I were to attempt re-writing your idea, I would say:

When the protocol behavior specified in Section 4.2.3.4 is implemented and
TaskReporting=FastAbort is operational on a session, then the FastAbort
functionality is supported/exhibited on that session by the implementation.

>a.	Section 4.2.3.5: If an iSCSI target implementation is capable of
supporting
>TaskReporting=FastAbort functionality (Section 13.23).
 
"Supporting" = implemented and negotiated to "TaskReporting=FastAbort".
Note that if even if an implementation has implemented this functionality
and wants to use it, it will not be able to negotiate it on any session if
it's only communicating with a bunch of RFC 3720 implementations.  

 >b.	Section 4.2.2.3.3: Whenever the TaskReporting key (Section 12.23
"Task Reporting") is negotiated to ResponseFence or FastAbort for an iSCSI
session.
>c.	Section 4.2.2.3.4: Whenever the TaskReporting key (Section 13.23) is
negotiated to ResponseFence or FastAbort for an iSCSI session.

I don't see an issue.  Note that the discussion in these two sections is on
response fencing, not fast abort per se.

>4.	...prefixed with some text like "Only when the
TaskReporting=FastAbort functionality is supported..".

Please see the above responses.  Your wording is vague for me, as I don't
comprehend a standalone notion of "supporting" a feature, other than
implementing a feature and negotiating the feature usage.   Current RFC 5048
requirement is that implementation is MUST, and negotiation is MUST (may
result in the feature to be used, or not).  So as such, there is already
ample flexibility for implementations.  

I think you are restating your earlier suggestion to loosen the "MUST
implement" requirement in RFC5048.  And I've already shared how I feel about
it, :)  I think it would be a step in the wrong direction, when faster error
recovery with multi-task aborts with large # of sessions/LUs/initiators is
even more critical, due to increased consolidation in data centers.  

At this point, I think it would be good to also hear from the rest of the
list.

Thanks.

Mallikarjun





>Thanks Mallikarjun! I would still like to confirm my understanding of what
you describe as "MUST implement, but MUST negotiate".
> 
>Here is my understanding:
> 
>1.	The implementation of TaskReporting is required and NotUnderstood
response for TaskReporting key is not permitted based on the following text
from RFC5048 and Section 6.2 of the consolidated draft:
>                 All keys defined in [RFC3720] MUST be supported by all
>            compliant implementations; a NotUnderstood answer on any of the
>            [RFC3720] keys therefore MUST be considered a protocol error
and
>            handled accordingly.
>2.	To enable FastAbort on a session, the negotiation of TaskReporting
key is required.
>3.	Only when the TaskReporting=FastAbort functionality is supported,
the protocol behavior specified in Section 4.2.3.4 is required to be
implemented and exhibited. There are several places in the spec that
indicate this type of conditionality.
>a.	Section 4.2.3.5: If an iSCSI target implementation is capable of
supporting
>TaskReporting=FastAbort functionality (Section 13.23).
>b.	Section 4.2.2.3.3: Whenever the TaskReporting key (Section 12.23
"Task Reporting") is negotiated to ResponseFence or FastAbort for an iSCSI
session.
>c.	Section 4.2.2.3.4: Whenever the TaskReporting key (Section 13.23) is
negotiated to ResponseFence or FastAbort for an iSCSI session.
>4.	If TaskReporting=FastAbort functionality is not supported, then the
protocol behavior specified in Section 4.2.3.4 is not required to be
implemented or exhibited. If this is true, then the first sentence in
section 4.2.3.4 is misleading and should be removed or prefixed with some
text like "Only when the TaskReporting=FastAbort functionality is
supported..".
> 
>Hemal
> 
> 
>-----Original Message-----
>From: Mallikarjun Chadalapaka [mailto:[e-mail]] 
>Sent: Friday, June 17, 2011 3:32 PM
>To: Hemal Shah; [e-mail]; [e-mail]
>Subject: RE: [storm] Plan for iSCSI work
> 
>Hi Hemal,
> 
>During RFC 5048 effort, I recall we settled on the "MUST implement, but
MUST
>negotiate" formulation after some list deliberation.  
> 
>With plain RFC 3720 semantics, there are some multi-initiator scenarios
>where multi-task aborts could essentially lead to target deadlocks, waiting
>on initiators on third-party sessions that may never respond.  With the
>"Clarified" semantics in RFC 5048, the third-party deadlocks aren't a
>problem anymore, but issuing initiator could still experience timeouts and
>escalate error recovery - depending on the number of LUs, sessions,
>connections and tasks affected with the issued TMF.    Escalated error
>recovery is usually something we want to avoid as it adds more
>delays/alerts/logs/failovers/failbacks etc.  Finally, "Updated" semantics
in
>RFC 5048 are meant to address this timeout problem - they permit targets to
>provide accelerated responses and allow them to deal with
>book-keeping/quiescing operations in a lazy fashion (which are the real
>culprits that trigger timeouts).  
> 
>Given all the above, the WG settled on FastAbort as a MUST-implement.  The
>"MUST negotiate" part then is to enable backwards-compatibility with RFC
>3720 implementations.  At this point, I'd rather not loosen the
Consolidated
>iSCSI draft requirement from what it is in RFC 5048.  I suggest we leave it
>the way it is.
> 
>Mallikarjun
> 
> 
> 
> 
> 
> 
>From: Hemal Shah [mailto:[e-mail]] 
>Sent: Thursday, June 09, 2011 3:57 PM
>To: Mallikarjun Chadalapaka; [e-mail]; [e-mail]
>Subject: RE: [storm] Plan for iSCSI work
> 
>Thanks Mallikarjun!
> 
>The way I read Section 4.1.3 of RFC5048 and Section 4.2.3.4 of the
>consolidated draft, the implementation of "FastAbort" feature is a MUST.
>From the first paragraph text from Section 4.1.3 of RFC5048, the first
>sentence mandates that all iSCSI implementation comply to this section but
>the second sentence say that the requirement in this section is conditional
>to the negotiation of "FastAbort" key. 
> 
>Protocol behavior defined in this section MUST be implemented by all iSCSI
>implementations complying with this document. Protocol behavior defined in
>this section MUST be exhibited by iSCSI implementations on an iSCSI session
>when they negotiate the TaskReporting (Section 9.1) key to "FastAbort" on
>that session.
> 
> 
>This is confusing. I think it will be better to remove the first sentence
>from 4.2.3.4 and clarify that the requirements in this section is
>conditional. I suggest rewording the first two sentences to below:
> 
>       iSCSI implementations may negotiate the TaskReporting (Section 9.1)
>key to "FastAbort" on an iSCSI session. Protocol behavior
>defined in this section MUST be exhibited by iSCSI implementations on an
>iSCSI session when they negotiate the TaskReporting (Section 9.1) key to
>"FastAbort" on that session.
> 
>Hemal
> 
> 
>-----Original Message-----
>From: Mallikarjun Chadalapaka [mailto:[e-mail]] 
>Sent: Friday, May 27, 2011 5:29 PM
>To: Hemal Shah; [e-mail]; [e-mail]
>Subject: RE: [storm] Plan for iSCSI work
> 
>Hi Hemal,
> 
>The text in this draft came verbatim from section 4.1.3 of RFC 5048.  There
>have been no changes in this area.
> 
>The new text (as well as the old text) requires the TaskReporting key to be
>negotiated to "FastAbort" before the multi-task abort semantics can be used
>on an iSCSI session.
> 
>Thanks.
> 
>Mallikarjun
> 
> 
> 
> 
>  -----Original Message-----
>  From: Hemal Shah [mailto:[e-mail]]
>  Sent: Friday, May 27, 2011 4:51 PM
>  To: Mallikarjun Chadalapaka; [e-mail]; [e-mail]
>  Subject: RE: [storm] Plan for iSCSI work
> 
>  Mallikarjun and David,
> 
>  I noticed one problematic item in the consolidated draft. This item is
the
>  requirement to support FastAbort. This feature was already defined in the
>  implementation guide, but it was optional. In this draft, it became a
>  required feature MUST - see in section 4.2.3.4.
> 
>  Do you know why the requirement was changed in the consolidated draft?
> 
>  I would like to keep the requirement optional as stated in the
>  implementation guide and not break backward compatibility.
> 
>  Hemal
> 
>  -----Original Message-----
>  From: [e-mail] [mailto:[e-mail]] On Behalf
>  Of Mallikarjun Chadalapaka
>  Sent: Thursday, May 26, 2011 6:50 PM
>  To: [e-mail]; [e-mail]
>  Subject: Re: [storm] Plan for iSCSI work
> 
>  Hi David,
> 
>  The list of changes you have called out are already done in the latest
>draft.  I
>  assume then that you are suggesting that the list itself should be
>included
>  in the next revision of the draft.
> 
>  Here's what I recall we have done so far:
>  1)  iSCSIProtocolLevel specified as "1", and added a related normative
>  reference to iSCSI-SAM draft
>  2)  Markers and related keys were removed
>  3)  SPKM authentication and related keys were removed
>  4)  Added a new section on responding to obsoleted keys
>  5)  Have explicitly allowed initiator+target implementations throughout
>the
>  text
>  6)  Clarified that implementations SHOULD NOT rely on SLP-based discovery
>  7)  Added UML diagrams, and related conventions
> 
>  The above is of course in addition to consolidating the different RFCs,
>and
>  making the related editorial changes.
> 
>  Thanks.
> 
>  Mallikarjun
> 
> 
> 
> 
>    -----Original Message-----
>    From: [e-mail] [mailto:[e-mail]] On Behalf
>    Of [e-mail]
>    Sent: Friday, May 20, 2011 2:49 PM
>    To: [e-mail]
>    Subject: [storm] Plan for iSCSI work
> 
>    I thought I'd offer some advance planning/warning on this, as the
>    consolidated iSCSI draft (draft-ietf-storm-iscsi-cons) is large (over
>300
>    pages).  The current plan is to run a simultaneous WG Last Call on both
>  this
>    draft and the new features draft (draft-ietf-storm-iscsi-sam) starting
>in
>  mid-
>    June (probably the week of June 13, after I get back from a badly
needed
>    vacation).  That WG Last Call will run longer than the typical 2-week
>time
>    period, due to the total size of the drafts, but will end by July 5th
at
>the
>    latest so that the status of the drafts and the next steps are known
>prior
>  to
>    the T10 (SCSI standards) meetings during the week of July 11.  As July
>11th
>    is also the draft cutoff deadline for the Quebec City IETF meetings,
>revised
>    draft versions may not show up until that meeting week (week of July
>    24th).
> 
>    This is also a good point to announce that the storm WG will meet in
>    Quebec City.
>    I've only requested a 1-hour session, as we get most of our work done
on
>    the mailing list.  Among the items for that meeting will be figuring
out
>  what
>    to do with the RDMA extensions draft (despite its name,
>draft-ietf-storm-
>    rdmap-ext-00, it's not currently an official work item for the storm
>WG).
> 
>    One thing that's missing from the consolidated iSCSI draft (and is a
>reason
>    why we're going to need a -03 version) is the changes that it makes to
>the
>    RFCs that it consolidates.  Off the top of my head, the major changes
>are:
>         - Removal of SPKM authentication
>         - Removal of the Marker appendix
>         - Removal of the SHOULD requirement for SLP implementation.
>    Have I missed anything significant?  The summary of this will need to
be
>    added to that draft.
> 
>    WG Last Call will be an opportunity (in fact the final opportunity) to
>  discuss
>    whether anything else should be removed from iSCSI, but there's no need
>    to wait
>    - I encourage people to review both drafts and post comments whenever
>    they can.
> 
>    In parallel, work will get started on any iSCSI MIB changes that are
>needed.
>    So far, I only see one MIB change - the iSCSIProtocolLevel from the new
>    features draft needs to be added to the MIB, probably with a structure
>    analogous to the iSCSI version support that's already in the MIB.
> 
>    Thanks,
>    --David (storm WG co-chair)
>    ----------------------------------------------------
>    David L. Black, Distinguished Engineer
>    EMC Corporation, 176 South St., Hopkinton, MA  01748
>    [phone]             FAX: [phone]
>    [e-mail]        Mobile: [phone]
>    ----------------------------------------------------
> 
>    _______________________________________________
>    storm mailing list
>    [e-mail]
>    https://www.ietf.org/mailman/listinfo/storm
> 
>  _______________________________________________
>  storm mailing list
>  [e-mail]
>  https://www.ietf.org/mailman/listinfo/storm
> 
> 
> 
> 
> 
> 
> 
> 
>