Joint T10/T11.3 Activity Working Group Minutes 16 Feb 2000
stewart_wyatt at hp.com
Thu Feb 24 16:30:53 PST 2000
* From the T10 Reflector (t10 at t10.org), posted by:
* "WYATT,STEWART (HP-Boise,ex1)" <stewart_wyatt at hp.com>
Joint T11.3/T10 Activity Working AdHoc Group T11/00-118v0
Huntington Beach, CA
February 16, 2000
Stewart Wyatt, HP, Secretary
1. Introductions: Dale LaFollette
Facilitator Dale LaFollette called the meeting to order and had the
participants introduce themselves.
2. Approve this agenda: T11/00-003v1 - Dale LaFollette
The agenda was approved but the items were reordered to postpone resolving
the FCP-2 letter ballot comments until all of the other agenda items had
3. Approve 01/13/00 minutes: T10/00: T10/00-130r0 - Bob Snively
Stewart Wyatt reviewed the minutes of last months meeting, which were
approved without comment. Stewart also thanked Bob Snively for his taking
4. Review old action items:
#1. Dave Peterson: Propose reasonable timer values, including Dave
Baldwin's proposed reduction of E_D_TOV. Completed.
#2. Charles Binford: Proposal to notify initiators of cleared commands to
be taken to the SCSI working group. In progress.
#3. Dale LaFollette: Prepare agenda for January meeting in Australia.
#4. Bob Snively: Facilitate and provide minutes for the January meeting in
#5. Bob Snively: Move the diagrams of Annex C into clause 11, making them a
normative example. Elected not to make the change for editorial reasons.
#6. Bob Snively: Post the four byte multiple fixed block length and error
recovery decision to the reflector. Completed.
#7. Bob Snively: Include revised Annex E (T11/99-340v3) in the next revision
of the FCP-2. Completed in FCP-2 revision 4. This item is still open and the
subject of a letter ballot comment.
#8. Bob Snively, Carl Zeitler, and Dave Peterson: Review the impact of
adding out-of-order delivery to the FCP-2. Schedule a review in the February
meeting. Carl Zeitler made the presentation, which was accepted.
#9. Paul Suhler will prepare a proposal for addressing the problems in the
LOCATE and READ POSITION commands for consideration at the February meeting.
#10. Bob Snively will include a comment requesting the removal of Annex J in
the Sun letter ballot comments. Completed
+++ Joint T10/T11.3 +++
5. New Business
5 A. Carl Zeitler FCP-2 Out Of Order Presentation T10/00-137r0
Carl duplicated the error recovery diagram examples in Annex D of FCP-2 Rev
4 and updated each of the cases to accommodate out-of-order frame reception.
Carl stated an assumption of his that performance is not an issue with error
Carl started with Class 3 examples, commenting that the changes to Class 3
were relatively minor. Carl provided a new definition of REC_TOV, which was
equal to half of R_A_TOV. To briefly summarize the Class 3 examples, there
are two significant cases. The first is where a frame was assumed lost on a
"one way" trip. Error recovery is initiated after REC_TOV. An example would
be a lost data frame on a write. The initiator waits REC_TOV then sends an
REC. The second case is when the frame was assumed to be lost on a "round
trip" then two REC_TOV waits (equal to R_A_TOV) are required. An example
occurs when a FCP_CMND only requires a FCP_RSP and the FCP_RSP is lost. The
initiator waits REC_TOV before sending the REC, and then waits another
REC_TOV for the original response before initiating error recovery.
Matt Wakeley noted that there is no requirement for when the next sequence
is sent in FC-PH. The recovery timers assume that the reply is sent
instantaneously and the delays are all in the fabric, but that is not a
requirement of the standard.
Carl noted that Class 2 requires more changes than Class 3. He also claimed
he was using Class 2 in the "classic" sense. Carl's approach followed the
procedure described in FC-PH. It was soon apparent that the participants
were not conversant with the standard procedure and a lot of discussion and
references to FC-PH occurred. Carl admitted that he had to do some reviewing
himself. Carl defended the common error recovery approach between the
classes of service already established by this committee. His proposal
requires ABTS to abort sequences (not exchanges) and requires the use of
One issue that surfaced was that Sequence ID is not a qualifier of Sequence
Count. In recovering from an error, a new Sequence ID must be used and the
Sequence Count must be incremented from the value used in the ABTS. Bob
Snively noted that this should have been required in the in-order case as
Carl observed that the RRQ was optional for the target when an ACK was lost.
This prompted a long discussion. Bob Snively thought that the RRQ was
necessary to prevent the Sequence Recipient from reusing the OX-ID of the
lost ACK. It was resolved that the Sequence Initiator must retire the RX-ID
until the RRQ expires, but that there is no obligation on the Sequence
Recipient. Matt Wakeley was concerned that without establishing an RRQ that
the EE_Credit count would be corrupted if the ACK did appear later. A
resolution of this dilemma was to recover the credit when the RRQ was
established and not claim credit it for the ACK if it does appear later.
There was considerable discussion over these issues and numerous references
back to FC-PH.
Another issue that excited considerable discussion was whether a BA_RJT is
allowed for an ABTS. Carl noted a special case of a lost command. If the
first frame of an exchange is lost, the Exchange Responder cannot reject the
ABTS even though it has unfamiliar with the Exchange. The ABTS must be
accepted to set up the RRQ in case the command appears later. There was some
concern that issuing a BA_RJT needs to be prohibited since it does not set
up a recovery qualifier.
After Carl had presented his slides the group reviewed the impact. Bob
Kembel noted that the FCP-2 error recovery proposal allows terminating
Exchanges in a manner not documented in FC-PH. Stewart Wyatt asked how a
device that only supports in-order command reception would signal that
capability. The answer is that in-order requirements are given to the fabric
in the FLOGI. The Class 3 case requires the changes to the SEQ_CNT and
SEQ_ID, noted earlier, though that change is independent of the in-order
issue. Class 3 also requires changing REC_TOV.
A motion was forwarded for the groups approval by Carl Zeitler and seconded
by Bob Snively: Do we want to use Carl's proposal as amended (which supports
both in and out of order delivery) as the only error recovery procedure for
FCP-2? On a company vote the measure passed 14 for, 3 against, 2 abstaining.
5 B. Data Transfer Integrity T11/00-021v0 Dale LaFollette
Dale LaFollette made a presentation that he had prepared with Matt Wakeley.
Dale observed that the PCI bus used internally to some initiators and
targets is the weak link in protecting the data during a transfer since
there is only 1 bit of parity covering 32 bits of data. (Ed Gardiner noted
that some designs do not even check the parity.) Dale's proposal included
several approaches to protecting data over the entire transfer but not
including the actual storage media.
The most promising proposal was to have the target create a checksum, which
it returns to the host in the response. The host also creates a checksum,
which it compares to ensure that the data transferred without error. This
approach works well for serial protocols like Fibre Channel, but not for
George Penokie vigorously opposed the proposal, noting that "the paranoid
have already solved this problem". His reference was to vendor unique
solutions and he felt that there is no need for a standards solution. Bob
Snively countered that more people are becoming paranoid. Joe Breher
expressed concern with the protocol attempting to solve internal device
problems. Ed Gardiner expressed concern about obtaining host side support
for this feature.
Dale called for a straw poll, "should this proposal be continued to be
included in FCP-2?" The results were 6 for, 5 against with a large number
5 C. Read FCP_XFER_RDY T11/00-067v0 Dave Peterson
Dave's proposal was to add a FCP_XFER_RDY from the initiator for reads. FCP
had a transfer ready for reads that came from the target, which was
obsoleted in FCP-2. Normally the host sends commands for space that it has
available and does not need this level of flow control. Dave justified his
proposal by noting that some times an intermediate device such as a router
does not know the available buffer space in the host and wants to use the
transfer ready for flow control. This proposal led to a long discussion. The
longest argument about target's role in controlling data transfers. Charles
Binford noted some precedents for this type of behavior. Bob Snively
expressed concern about how it would affect error recovery. There was also
some paranoia expressed about the effect this would have on performance and
the risk of deadlock. Finally Dale LaFollette called a straw poll asking if
the group would support continued effort to develop this proposal. The
results were for continuing 5, against continuing 7, abstaining 6.
Matt Wakeley expressed frustration with those who voted against further
investigation of optional proposals, referring to this proposal and the
previous one. The reluctance comes partly from target manufacturers who fear
that optional features would become mandatory for them.
+++ T11.3 +++
6. T11.3 New Business:
+++ T10 +++
7. T10 New Business
7 A. SSC Public Review Comment
An SSC public review comment was received from Paul Suhler requesting
restoring the SET CAPACITY Command, which was in early versions of SSC but
not in the final version. There was some procedural discussion between
George Penokie and Dave Peterson. As the discussion continued the history of
this command became more suspect. Checks in other documents indicated that
the command had never actually existed. The conclusion was that it was only
a proposal that had never been finished. Paul wanted to have the command
revived to support reducing the capacity of a large tape. He promised to
consider using a partition size or some mode page as an alternative to using
7 B. Response to PRLI/ACC with both I/T bits set. Neil Wanamaker
This issue had surfaced in the FC_MI group. During interoperability testing
it was found that some hosts and targets will fail to complete PRLI is the
other party sets both the host and target bits. Examples of this case
include disks implementing the XOR command and tapes implementing the COPY
command. A specific issue is when a target, which can also function as an
initiator, attempts to send a PRLI to an initiator. Some initiators will not
complete the PRLI unless they initiated it. Neal is working on the
appropriate response in each of these cases.
7 C. SSC Large Block Addresses T10/00-135r0 Paul
Paul noted that with the large capacity of emerging tape drives there is a
need to increase address space in commands. Of the methods proposed, using
16 byte commands appears most acceptable. Paul marked up his proposal with
inputs from the group, which included the affected commands.
+++ Joint T10/T11.3 +++
8. FCP-2: T10 Working Drafts FCP2R04 and Letter Ballot Comments T10/00-150r0
The letter ballot review had been put off until last. Bob noted that there
were 1350 letter ballot comments to review. Bob said he planned to go
through them and answer the editorial comments and the obvious technical
comments. He would review the more difficult or controversial comments with
the group. The attendance had diminished as the evening wore on. A reception
was in progress in a near by room. As it was quite late, the meeting was
adjourned without starting the letter ballot review.
9. Next Meeting Requirements - Dale LaFollette
The next meeting is being held in Dallas, Texas, at the Crowne Plaza Suites
sponsored by Texas Instruments. The main agenda item will be reviewing the
FCP-2 letter ballot comments. The meeting was originally scheduled for
Tuesday March 7, which conflicted with the Parallel SCSI Working Group.
After some discussion the meeting time was changed to Monday March 6. The
meeting will start at 9 AM and is scheduled to run until 6 PM.
10. Review New Action Items: Stewart Wyatt
#1 Carl Zeitler - Update the Out-Of-Order proposal drawings in T10/00-137r0.
#2 Carl Zeitler - Determine the correct response to a lost ACK: BA_RJT or
#3 Bob Snively - End exchange cases in FC-PH does not include the class 3
case of lost FCP_CONF. Needs to be added to FC-FS. Check for other cases.
#5 Paul Suhler - SSC letter ballot comment requesting restoring the Set
Capacity command. Paul will investigate some other means of limiting the
available capacity of a large tape.
#6 Neil Wanamker - Revise proposal defining behavior with both target and
initiator bits set in PRLI.
#7 Paul Suhler - Revise proposal for Large Block Addresses in the SSC.
Old Action Items
#8. Charles Binford: Proposal to notify initiators of cleared commands to
be taken to the SCSI working group.
#9. Bob Snively: Include revised Annex E (T11/99-340v3) in the next
revision of the FCP-2. Completed in FCP-2 revision 4. This item is still
open and the subject of a letter ballot comment.
#10. Bill Martin: Bob Snively requests that Bill Martin review the
Out-Of-Order proposal looking for corner case problems.
11. Adjournment: Dale LaFollette
Dale LaFollette STK Stewart Wyatt HP
David Peterson STK Arlan Stone Unisys
George Penokie IBM Ralph Weber ENDL
John Lohmeyer LSI Logic Matt Wakeley HP/Agilent Tech
Dennis Moore KnowledgeTek Charles Monia Adaptec
Neil Edmunds Xyratex Predraig Spasic HP
Paul Suhler Seagate Horst Truestedt TrueFocus
Jim Coomes Seagate Curt Ridgeway LSI Logic
Bob Snively SUN Edward A. Gardner Ophidian Designs
Robert Reynolds Crossroads Systems Pak Seto Quantum
Bret Ketchum CNT Kumar Malavalli Brocade
Bob Kembel Connectivity Sol. Craig Stuber JNI
Mike Fitzpatrick Fujitsu Jeff Stai Q Logic
Stephen O'Neil CMD Joe Breher Exabyte
* 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