Joint T10/T11.3 Activity Group Minutes for Dec 7, 1999

WYATT,STEWART (HP-Boise,ex1) stewart_wyatt at hp.com
Wed Dec 8 15:24:56 PST 1999


* From the T10 Reflector (t10 at t10.org), posted by:
* "WYATT,STEWART (HP-Boise,ex1)" <stewart_wyatt at hp.com>
*
Joint T10/T11.3 Activity Working Group AdHoc Meeting   T11/99-754v0
December 7, 1999, Reno Nevada

Stewart Wyatt - Secretary 

1 Introductions: Dale LaFollette 

The meeting started at 1PM. Dale had the group introduce themselves.

2 Approve this Agenda: T11/99-725v0 Dale LaFollette

Approved with minor additions.

3 Approve 11/02/99 Minutes: T11/99-679v0 Stewart Wyatt

Approved

4 Review Old Action Items: Stewart Wyatt

#1. Charles Binford, LSI Logic. Refer default E_D_TOV issue to FC_FS. Prefer
a 2 second to 10 second for point to point connections. Reassigned to Dave
Peterson

#2. Charles Binford, LSI Logic. Propose a new RESP_CODE 0x06: Command
cleared by another initiator. Take new status code to SAM-2. Ongoing

New Action Items

#1 Dale LaFollette, STK. Prepare agenda for the January T10 meeting in
Australia - Ongoing

#2 Bob Snively, SUN Microsystems. Facilitate and take minutes for the
January 2000 Joint Activity meeting in Australia. - Ongoing

#3 Bob Snively, SUN Microsystems. Investigate whether it is appropriate to
include the discovery ELS (RTIN and RNID) in the list of ELS to be accepted
before completing login. - Moved  to  FC-FS

#4 Bob Snively, SUN Microsystems. Move the diagrams of Annex C into clause
11, making them a normative example. - Next FCP revision

#5 Bob Snively, SUN Microsystems. Verify the multiple of 4 byte block length
change is acceptable for both the SSC-2 and FCP-2 by posting a proposal to
the reflector and bringing it up at the SCSI plenary. - Plenary presentation
completed but not the reflector item

#6 Dave Peterson and Dal Allan. Review the current FCP-2 for limitations to
out-of-order error recovery implementations. - Carl Zeitler presentation
this meeting, additional action items continue the issue

#7. Bob Snively. Identify the problems with allowing unlimited process
associators and prepare a review. - Bob made a counter proposal presented
over the reflector and at this meeting

#8 Dave Peterson. Review Annex E SCSI Device Discovery Procedure, to
determine what the differences are for target and initiator requirements
are. - Completed 

+++ Joint T10/T11.3 +++

5 FCP-2 T10 working drafts FCP2R03 and Change comments T10/99-325r1 Bob
Snively

Bob reviewed the new revision of his change document.  There were only a few
issues that generated extended discussion.

Out of Order issue: Bob invited Carl Zeitler to review his presentation. He
noted that the in-order requirement compromises complex fabrics.  Stewart
Wyatt observed that in a backup environment, allowing out-of-order would
require additional buffering for tape drives which operate sequentially. He
noted that out-of-order requirements would reduce the costs of fabrics while
increasing the cost of peripherals. 

Carl noted that time out values are critical in out-of-order error recovery.
Currently REC_TOV is defined as ED_TOV + 1 second. With these values in an
out-of-order environment, the target could receive a REC before it received
the command. The solution is to make the REC_TOV large enough to guarantee
that the frames are not still in the fabric. 

Carl presented a series of overheads that used ladder diagrams to
demonstrate the various error recovery cases for both Class 2 and 3. The
first set for commands, transfer readies and responses were relatively
straight forward. There was some confusion over Carl's use of ABTS. Carl was
using the ABTS to recover status on the sequence and set up a recovery
qualifier, not to abort the exchange.  In Class 3 Carl used RA_TOV/2 for the
timeout value instead of REC. Carl questioned the value of REC in Class 2.

The data transfer examples had more complexity. Some issues were raised.
Carl's overheads indicated the detection of a sequence error when the frame
was lost which is only true with in-order error detection. Out-of-order does
not detect an error until the timeout expires. The overheads also showed
ABTS being used incorrectly. Another concern was expressed over the ability
of an HBA to detect that a frame has been transmitted to start a timer since
these events occur at different levels in the HBA. Matt Wakeley noted that
ED_TOV will expire at the sequence recipient before the sequence initiator
will detect that an error occurred. 

There was a lot of discussion about recovering lost ACKs in Class 2. In an
ACK 1 form, the only ACK that must be recovered is the last one of the
sequence. An RRQ is always required to avoid confusion if the missing ACK
pops out later.
 
Carl also raised the issue of aborting a failing sequence in Class 3 during
an exchange that is transferring a large amount of data. This group had
previously discussed this issue and the feeling was that with the high
transfer rate, the delay would not be severe enough to justify action.
Nevertheless, Dale LaFollette has a proposal to use SRR to get the target's
attention that will be an annex in the next revision of the FCP. 

After considerable discussion, Dale LaFollette asked the group what they
thought. Dal Allan and Ed Gardiner both thought the issue was very
manageable, being largely an issue of timeouts, and thought that it should
be included. Bob Snively thought it would require more extensive code
structures in the host and be less robust. Dale asked Bob how long he
thought it would delay the FCP-2 to add it. After some discussion, the
answer was six months.

Stewart Wyatt expressed concern that delaying the FCP would reduce support
for developing drivers and applications for use by early implementations,
which will only support in-order delivery. He was concerned that this would
delay the market for Fibre Channel tapes. Stewart was also concerned about
making a clear distinction in the documentation between in-order and
out-of-order systems. Jim Coomes asked how an in-order and out-of-order
peripheral would be distinguished. A means will need to  be identified.

Stewart proposed completing the FCP-2 as is and starting a FCP-3 that
supports out-of-order. George Penokie seconded the motion. After further
discussion, the motion was withdrawn to allow Bob and others to better
access the work involved and define the impact. 

Process Associators: Bob Snively had created a counter proposal to replace
the need for process associators. Bob noted that targets could use LUNs for
separate representations. Other groups are developing access control. The
only missing piece is providing a means of placing multiple initiators
behind a single port. The solution Bob is proposing is to have a separate ID
with a distinct world wide name for each initiator. He thought that his
approach would have the least impact on switches. 

Jeff Stai expressed a concern that the scope of this discussion be strictly
limited to FCP. (FS has a separate agenda for process associators.)

While the proposal was positively received, Dal Allan and Matt Wakeley
questioned the number of addresses available to define the initiators. Dal
noted that ten years ago 8 bytes was considered necessary. Carl Zeitler
thought that a single byte (256 initiators) would be adequate. Later Carl
expressed concern about the numbers required for Infiniband support. Ed
Gardner reassured Carl that these numbers were adequate.

Dal thought each application could require a unique initiator to support
zoning, assuming that each application would need its own zone. Bob Snively
argued that process associators had not been involved with zoning.
 
Finally the group came to a conclusion. Bob's proposal would be the solution
for FCP. (Process Associators would be a FC-FS function to appear in
FC-SW-2.) The policy will be that Process Associators does not participate
in task identification. It was suggested that the group endorse this
decision. The lone dissenter was Matt Wakeley over concerns that the
available number was too small. (This concern may provide an incentive
elsewhere to develop header to increase the address space - and these may
look like process associators!)

Incorrect use of recovery abort: Bob noted that Charles Binford's proposal
(discussed below) reduces the number of ambiguous changes and the incorrect
usage of recovery aborts. A mode bit enables the features of Charles'
proposal. Bob though that this function should be required. Bob expects that
ABTS will be used to cancel any remaining ambiguous exchanges, even
requiring this functionality from a Class 3 target.

EMDP bit: Bob had required the EMDP bit to be set for SRR error recovery.
Matt Wakeley expressed concern that setting this bit implies support of
out-of-order delivery, which is not what was intended. Charles Binford
proposed that REC/SRR error recovery be allowed without setting this bit.
This proposal was accepted.  

6 FCP-2/FC-GS-3 Update T11/99-710v0 Dave Peterson

Dave has made a proposal that was accepted by FC-GS-3 working group to
assign four bits for FC-4 specific usage. The bits will indicate whether the
device supports SCSI target and/or initiator behavior. There was some
discussion about this capability. 

7 REC/SRR Questions T11/99-728v0 Neil Wanamaker

Neil's slides were reviewed during FCP-2 discussion. 1st and 3rd slides
accepted. The 2nd slide needs review to resolve an ordering issue.

8 Task Management Questions Dale LaFollette

Dale noted that the FCP-2 ambiguously describes the targets response to a
task management function as either accepted or performed. Ed Gardiner,
referring to SAM, noted that these terms are the same. Dale asked how a
device should respond before completing a target reset. The answer is to
respond with a busy. 

+++ T10 +++

9 SSC: T10 Working Drafts SSC-R22 and change document T10/99-228r4 Dave
Peterson

No news, waiting for the public review to close.

+++ New Business +++

10  CMDS Cleared 99-722v0 Charles Binford

This proposal was originally made to this group. Charles was sent to present
it to the SCSI working group. This agenda item was an update. Charles made a
presentation to the SCSI group, which was positively received. He is
planning on working with Ralph Weber to make a formal proposal. 

Charles has two objectives in this proposal: The first is to notify other
initiators that their exchanges were cleared by a TMF from another
initiator. The second is to quickly clean up aborted I/Os to remove
ambiguous exchanges. 

The solution is to introduce a new SCSI status 'CMD Cleared', sent to all
initiators with outstanding aborted commands including the one who cleared
them. No ordering requirements are assumed. Other causes of clearing
commands (PLOGI, LOGO, etc.) are unchanged. Charles prefers that targets
send a PRLO to all initiators if it receives a TPRLO. The Control mode page
will enable this ability.
 
11 Next Meeting Requirements: Dale LaFollette

Dale asked how much time would be required for the meeting in two months.
Dave Peterson was asked if he needed time for reviewing the SSC after the
public review is completed. He didn't think so. Bob Snively hopes to have
completed the letter ballot on the next revision of FCP-2 by then and
requested three hours minimum.

12 Review New Action Items: Stewart Wyatt

#1. Dave Peterson. "Reasonable" timer values proposal, including Dave
Baldwin's proposed reduction of E_D_TOV.

#2. Charles Binford. Command cleared proposal for the SCSI working group.

#3. Dale LaFollette. Prepare agenda for January meeting in Australia.

#4. Bob Snively. Facilitate and provide minutes for the January meeting in
Australia.

#5. Bob Snively. Move the diagrams of Annex C into clause 11 making them a
normative example.

#6. Bob Snivley. Post the four byte multiple fixed block length and error
recovery decision to the reflector.

#7 Bob Snively. Include revised Annex E (T11/99-340v3) in the next revision
of the FCP-2. 

#8 Bob Snively, Carl Zeitler, Dave Peterson: Review the impact of adding
out-of-order delivery to the FCP-2. Schedule a review in the February
meeting. Have Bill Martin review the proposal.

13 Adjournment: The group adjourned at 5:40 PM

Attendance List:

Dale Lafollette	StorageTek	Stewart Wyatt	HP
David Peterson	STK	Roger Cummings	DPT
Joe Breher	Exabyte	Neil Wanamaker	Crossroads Systems
John Scheible	IBM	George Penokie	IBM
Bob Snively	SUN	Vit Novak	SUN
Damian Bannon	SSL	Pak Seto	Quantum
Chuck McKnett	JNI	Steve O'Neil	CMD
Matt Wakeley	HP/Agilent	Bill Martin	Gadzoox Networks
Jim Coomes	Seagate	Charles Binford	LSI
Danny Ybarra	TI	Carl Zeitler	Compaq
Gene Milligan	Seagate	Arlan Stone	Unisys
Dave Guss	TI	Mark Hamel	Compaq
Paul Suhler	Seagate
*
* 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 mailing list