Minutes of SSC/SMC WG Meeting 13-Mar-97 (97-154r0)

Erich Oetting erich_oetting at stortek.com
Thu Mar 20 15:56:21 PST 1997


* From the SCSI Reflector (scsi at symbios.com), posted by:
* erich_oetting at stortek.com (Erich Oetting)
*
Accredited Standards Committee         Doc. # X3T10/97-154-r0
X3, Information Technology

To:       Members of X3T10

From:     Erich Oetting

Subject:  Minutes of SSC/SMC Working Group Meeting
                 San Diego, CA, Mar 13, 1997

                          Agenda

1. Opening Remarks

2. Approval of Agenda

3. Attendance and Membership

4. SMC Topics

4.1 Read Element Status and Reservation Conflicts. (96-267-r1)

4.2 SMC Device Identification (97-148-r0)

5. SSC Topics

5.1 PLDA Tapes (Doc TBD).

6. Other Topics.

7. Meeting Schedule.

8. Adjournment.


                     Results of Meeting

1.    Opening Remarks

Erich Oetting, the SMC Technical Editor, called the meeting to order at
9:10 a.m., Thursday, March 13, 1997.  He thanked Skip Jones and QLogic for
arranging and hosting the meeting.

As is customary, the people attending introduced themselves and a copy of
the attendance list was circulated.

The draft agenda was approved.

3.    Attendance and Membership

Attendance at working group meetings does not count toward minimum
attendance requirements for X3T10 membership.  Working group meetings are
open to any person or organization directly and materially affected by
X3T10's scope of work.

The following people attended the meeting:

       Name        Organization          Email Address
--------------    -------------       -------------
Connie Kephart     Exabyte           conniek at exabyte.com
Erich Oetting      Storage Tech.     erich_oetting at stortek.com
Ralph Weber        Symbios Logic     roweber at acm.org
George Penokie     IBM               gop at rchump3.vnet.ibm.org
Gene Milligan      Seagate           gene_milligan@
                                       notes.seagate.com
Bob Snively        Sun               bob.snively at sun.com
Charles Monia      Digital           monia at shr.dec.com
Dal Allan          ENDL              dal_allan at mcimail.com
Doug Hagerman      Digital           hagerman at starch.enet.dec.com
Dan Davis          Overland Data     ddavis at overland.com
David Cressman     Quantum           dcressma at tdh.qntm.com

Dan Total of 11 People Present

4. SMC Topics

4.1 Read Element Status and Reservation Conflicts. (96-267-r2)

Rob will bring a revised proposal to the May working group meeting.

4.2 SMC Device Identification (97-148-r0)

Erich presented his proposal to obsolete the SCSI Bus Address, LUN, ID
valid, LUN valid, and Not Bus fields and replace them with Identification
Descriptors. Some concern was expressed about obsoleting the SCSI Bus
Address etc. fields.  Erich will revise the proposal to leave these fields
in and add wording that prohibits using them when the SCSI Bus Address or
LUN numbers are too large to fit in the field.  The revised proposal
(97-148-r1), was recommended for adaption by the plenary.

5. SSC Topics

5.1 SSC PLDA Tapes (Doc 97-xxx-r0).

Doug talked about his paper on PLDA tapes and error recovery. His proposal
adds a confirmation of receipt of Read data from the Target by the
Initiator. Changes to PLDA and a minor addition to FCP would be required.

PLDA error recovery for disks assumes that error recovery is a simple
matter of resending the failed command.  The decision to send the command
is based on some error indication, or simply a timeout due to status not
being received in a few seconds.  Tape devices can't use the same technique
due to the long timeouts required, and the fact that tape position must
somehow be reestablished before the command can be reissued.

Doug expressed the wish to use existing tape host software transparently
over a Fibre Channel interface, with the  error recovery performed below
the command level. David said what we really needed was an equivalent to
Restore Pointers to allow recovery from data transmission errors.

Discussion then centered on the expected error rate on an arbitrated loop.
It was generally agreed that loops may exhibit a higher error rate than
current well designed parallel SCSI systems. Arbitrated loop glitches can
also be expected when devices are inserted or removed from a loop. It was
generally
agreed that the worst case numbers were apocryphal, but that
the error recovery problem still must be solved.

Dal voiced the opinion that all the tools to do error recovery are
currently available, but in some cases prohibited by the PLDA profile.  If
drives use incrementing sequence numbers, a missing sequence can be
detected. The exchange status can be polled to detect what part of an
exchange failed.  It was pointed out by others that exchange status has not
implemented in most existing designs.

Block size restriction were discussed as a method of getting around
problems that occur when a drive starts writing before the complete record
is received. It was pointed out that SSC currently requires SCSI tape
devices to restore position if a complete block is not received, even if
the drive started writing the block to tape before the failure. This
eliminates the need to restrict block size.

There was general agreement on some points during the meeting:

Error recovery should happen below the command layer, or at least invoke
command layer recovery that is already present.  There was a lot of
disagreement about what error recovery is actually present at the SCSI
command layer.  (Do current SCSI tape drives just give up if a transport
layer problem occurs?)

Tape media errors were deemed to be outside the scope of this discussion.
Also, the current buffered model, with a queue depth of one used by most
existing SCSI tape drives will be retained. This does not preclude using
command queuing on future tape devices.

Ideally, errors that occur in the transport of commands, data and status
will be recovered without invoking the SCSI command layer. A method of
requesting retransmission of data equivalent to the Restore Pointers
message in parallel SCSI is desired.  FCP_XFR_RDY can be used to retry a
target detected data errors for a write command.  However, there is no hook
that allows an initiator to request a retransmission of read data from a
target. Sending a sequence or logical block (one or more sequences) again
was preferred to attempting to recover at the frame level.

Where possible, direct detection of errors is preferred to timeouts. Errors
that can't be recovered should be reported using in a fashion that will
invoke existing SCSI command level error recovery.  There was some
discussion of what, if any error recovery is done today on parallel SCSI.

Changes for tape support to PLDA and FCP should be compatible with existing
silicon. More input about the capabilities of current FC silicon will be
needed by the working group.

Class 2 may have been desired, but is not needed to solve the problem.

Target documents for these changes will be PDLA-2 and FCP-2.

6.   Other Topics

No other topics were discussed.

7.    Meeting Schedule

The next meeting of the SSC/SMC Working Group will be next May in Natick
Massachusetts. Actual meeting time will be determined by the X3T10 Plenary.

8.    Adjournment

The meeting was adjourned at 12:15 pm on Thursday, March 13, 1997.



*
* For SCSI Reflector information, send a message with
* 'info scsi' (no quotes) in the message body to majordomo at symbios.com




More information about the T10 mailing list