Minutes of RAID working group 01/11/94

George Penokie GOP at rchvmp3.vnet.ibm.com
Wed Jan 12 16:03:40 PST 1994

 Minutes for RAID working group 01/11/94
 Accredited Standards Committee
 X3, Information Processing Systems
                                                    Doc. No.: X3T10/94-011R0
                                                        Date: January 11, 1994
                                                   Ref. Doc.:
                                                    Reply to: G. Penokie

 To:         Membership of X3T9.2 and X3T10

 From:       George Penokie

 Subject:    Minutes of RAID Study Group Meeting
             January 11, 1994


 1.    Opening Remarks
 2.    Attendance and Membership
 3.    Approval of Agenda
 4.    Report on last RAID Working Group
 5.    SCSI-3 Addressing (93-161r2)
 6.    SDA States and Types (93-191r1)
 7.    SDA Commands and Mode Pages (93-192r1)
 8.    SCSI Disk Array Model (93-140r3)
 9.    Dual-Controller Issues
 10.   Additional Device Models
 11.   Error Handling for SCSI Controllers
 12.   Action Items
 13.   Meeting Schedule
 14.   Adjournment

                               Results of Meeting

 1.  Opening Remarks

 George Penokie the RAID Study Group Chair, called the meeting to order at 9:10
 am, Tuesday, 11 January 1994.  He thanked Vitro for hosting the meeting.

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

 It was stated that the meeting had been authorized by X3T9.2 and would be
 conducted under the X3 rules.  However, the meeting will report to the newly
 formed X3T10 committee.  Ad hoc meetings take no final actions, but prepare
 recommendations for approval by the X3T10 task group.  The voting rules for
 the meeting are those of the parent committee, X3T9.2 and/or X3T10.  These
 rules are:  one vote per company; and any participating company member may

 The minutes of this meeting will be posted to the SCSI BBS and the SCSI
 Reflector and will be included in the next committee mailing.

 George stated that this is the 15th meeting of the RAID study group and the
 2nd joint RAID Advisory Board/RAID study group meeting.  The purpose of the
 group is to deal with interface issues related to using RAIDs.  The study
 group will assess the issues and then formulate a strategy for dealing with

 2.  Attendance and Membership

 Attendance at working group meetings does not count toward minimum attendance
 requirements for X3T9.2 membership.  Working group meetings are open to any
 person or company to attend and to express their opinion on the subjects being

 The following people attended the meeting:

                      RAID Study Group Meeting Attenders

                                                       Phone Number -or-
          Name          S       Organization         Electronic Mail Address
 ---------------------- - ------------------------- -------------------------
 Mr. Jerrie L. Allen      Amdahl                    408-944-3712
 Mr. Mark Skrondal        Amdahl                    408-944-3954
 Mr. John Woods           Amdahl                    408-944-5016
 Mr. Gerry Johnsen        Ciprico                   612-551-4055
 Mr. Steve O'Neil         CMD Technology            714-454-0800
 Mr. Bill Galloway        ComPaq                    713-374-6732
 Mr. Bob Solomon          Data General - Clarion    508-898-5014
 Mr. Doug Hagerman        Digital Equipment Corp.   hagerman at starch.enet.
 Mr. Ralph Weber          Digital Equipment Corp.   weber at star.enet.dec.com
 Mr. Doug Anderson        Distributed Proc. Tech.   407-830-5522
 Mr. Howard Grill         Formation                 609-234-5020
 Mr. Kevin Pokorney       Fujitsu                   303-682-6649
 Mr. Bill Hutchison       Hewlett Packard Co.       hutch at .boi.hp.com
 Mr. Andrew Hill          HI-Data                   617-937-6770
 Mr. Paul Boulay          Hitachi                   408-986-9770
 Mr. Giles Frazier        IBM Corp.                 512-838-1802
 Mr. George Penokie       IBM Corp.                 gop at rchvmp3.vnet.ibm.com
 Mr. Roy Kurz             Jaycor                    619-535-3117
 Mr. Charles Binford      NCR                       316-636-8566
 Mr. M.W. Jibbe           NCR                       316-636-8810
 Mr. Herb Silverman       Peer Protocols            714-476-1016
 Mr. Kim Le               STK - Ampen'f             303-673-7322
 Mr. Robert N. Snively    Sun Microsystems, Inc     bob.snively at eng.sun.com

 23  people present

  Status Key:  P   -  Principal    ** Status not applicable at this
               A   -  Alternate       meeting because X3T10 committee
               O   -  Observer        not yet formed.
               L   -  Liaison
               S,V -  Visitor

 3.  Approval of Agenda

 The agenda developed at the meeting was approved.

 4.  Report on last RAID Working Group

 Last RAID Working Group was held in San Diego, CA.  Several members of the
 RAID Advisory Board were present.  Minutes of the San Diego meeting are ready
 for publishing, but a document number has not yet been assigned.  When the
 document number is assigned the minutes will be published on the SCSI

 5.  SCSI-3 Addressing (93-161r2) Penokie

 George presented a revised version of the SCSI-3 Addressing proposal
 (X3T9.2/93-161r2).  George apologized for not distributing copies of the new

 George noted that the EXTENDED IDENTIFY message has been removed from the
 proposal.  The only 64-bit LUN addressing now proposed for SIP SCSI uses the
 translation table.  This is considered sufficient for all anticipated parallel
 SCSI implementations.  RAID systems that require larger address ranges must be
 implemented based on SBP, Fibre Channel, or another serial protocol.

 This generated questions about how the equivalent addresses are generated and
 used in the serial protocols.  Where is the VolSel bit equivalent?  It's the
 low-order bit of the Access Method field.  Bob Snively was called to describe
 Fibre Channel addressing maps into the addressing levels in 93-161r2.

 Bob noted that the 93-161r2 proposal cannot describe the full addressing
 capabilities available in Fibre Channel.  However, it can describe the add-
 ressing needs all but the most outrageous RAID configurations.  Particularly,
 it can address a large set of practical and useful configurations.

 The concern was raised that serial busses already have 64-bit LUNs.  So, the
 8-bit addressing level fields.  George will add words to 93-161.  In parallel
 SCSI, LUN values in addressing level fields are one-to-one with normal usage
 LUN values.  On the serial busses, some type of vendor specific mapping may be

 George also agreed to add examples that use pure IDENTIFY message formats to
 access storage array components.

 The issue of changing the mapping table resurfaced.  George said that no
 protections are provided beyond those mode page rules that apply to all SCSI
 mode pages.  Reserving LUN 0 is the only true protection.

 George indicated that r3 probably will be the last revision of 93-161.  He
 will make that revision and distribute it soon.  The group accepted George's
 proposal to never distribute r2, because r3 is almost the same as r2.  The
 next activity will be incorporating 93-161r3 in the SCSI-3 Controllers
 Commands (SCC) document.

 6.  SDA States and Types (93-191r1) Penokie

 George presented a new document defining states of an SDA and the types of
 devices needed within Disk Arrays (X3T9.2/93-191r1).

 A significant discussion took place about the use of peripheral device type
 for items such as fans and LED panels.  George has received written comments
 expressing displeasure with this usage and proposing alternatives.  The group
 reviewed a couple of options based on the SCSI-2 document.  Two fields of
 interest are the Peripheral Qualifier and the Device-Type Modifier.  Device-
 Type Modifier appeared preferable, because it is essentially reserved in
 SCSI-2 and because its 7 bits wide.

 George noted a couple of minor changes.  The changes generated no discussion.

 Doug Hagerman presented some comments on the SDA states in 93-191.  Doug
 contends that the state information in 93-191 is not sufficient to cover the
 real situations that occur in SDAs.  He thinks that reliable management of the
 SDA requires a careful definition of states, how the SDA gets into them, and
 how it gets out of them.

 Doug has prepared a document listing all the states that he can think of.
 Doug's document also separates the states into lists for the SDA, Redundancy
 Group, P-LUI, and Spare.  Doug's work is based on George's 93-191, and
 contains at least all the states found there.

 The group discussed the meaning of each state.  Examples were given to clarify
 meanings.  George will incorporate the states definition style into 93-191.
 He also will add those states in Doug's document that are not already present
 in 93-191.

 7.  SDA Commands and Mode Pages (93-192r1)

 George presented a new document defining new commands and mode pages needed to
 implement SCSI disk arrays (X3T9.2/93-192r1).  Recent revisions to the
 document bring it into alignment to the disk array model.

 The document does not define exact command formats.  It is more an outline
 that George is using to validate the model.  It lists the required inputs and
 outputs.  George will model the commands on existing SCSI commands.  However,
 he plans to avoid 6-byte command formats, because they do not have large
 enough parameter size byte counts.

 George noted the changes in the maintenance operations proposals.  Reporting
 functions for specific P-LUIs may require multiple command/response opera-
 tions.  While this increases the number of transactions required for some
 tasks, it greatly simplifies the operating model.

 No one could find where the state of the P-LUI is reported.  Efforts were made
 to locate it on two pages that were not duplicated.  George will either locate
 where P-LUI state is already reported or determine where reporting P-LUI state
 should be reported.

 A later discussion in the SCSI Disk Array Model reveled that George thought
 the state of the P-extent was the same as the state of the P-LUI.  But, what
 about fans?  Do they have P-extents?  At this point, George added a state
 field to the Report P-LUI service.

 Concerns were raised about rebuild times and priorities.  What commands does
 93-192 define for setting these kinds of dials.  Doug Hagerman suggested that
 mode pages are a better model for control knobs.  George asked that people in
 the group who have worked in this area bring in a proposal based on mode

 8.  SCSI Disk Array Model (93-140r3)

 George presented a revised version of the SCSI-3 Disk Array Model

 A half dozen minor wording changes in the glossary were not controversial.
 Doug asked for a clarification of the relationship between SCSI Disk Arrays
 (SDAs) and Disk Array Conversion Layers (DACLs).  Is the DACL a component of
 an SDA?  George said, "Yes."

 The Report P-LUI Geographic Location Service has been made separate from the
 Report P-LUI Service.  Further discussion resulted in the addition of a state
 field to the Report P-LUI Service return data.

 Questions were raised about how to get current state of a volume.  George
 described how one or more Report Service requests are used to learn the state
 of the volume.  More Report requests can be directed to the P-LUIs (based on
 the volume report), if more detailed state information is needed.

 A discussion of polling and log-based methods followed.  George requested that
 someone bring a proposal for a logging page.  This would cover polled/logged
 mechanisms for saving and reporting SDA state exceptions.

 The change-by-change review of the SDA Model produced no major comments.  Most
 of the comments were editorial in nature.

 9.  Dual-Controller Issues

 George described his concerns about adding dual controller definitions to the
 SCSI Disk Array (SDA) Model and/or SCSI-3 Controller Commands (SCC) document.
 George has no experiential knowledge of dual controller configurations.  So,
 he has no personal basis for leading standards development in that area.

 Also, George wants to expedite delivery of the information that already has
 been developed.  So, he is concerned about adding uncertainty to that

 George noted that dual controller work can be added in a second pass of the
 SDA Model and SCC.  This did not satisfy the gut-level interests of many
 people present.

 The group concluded that another member of the group would have to lead the
 dual controller discussion.  No one volunteered at the meeting.  On the other
 hand, no one proposed terminating any work on dual controllers immediately.

 10.  Additional Device Models

 Doug Hagerman presented the first draft of his document containing minimalist
 device models for things like power supplies and fans.  The intent is to make
 more components of a disk array addressable entities, so that they can be
 added to redundancy sets.

 The document contains device models for power supplies, fans, consoles, and
 caches.  Doug's goal is to test the validity of the draft device models.

 George stated a general philosophy that anything that is an addressable entity
 can be added to redundancy sets.  But, making a device type is not needed for
 making an addressable entity.  For example, P-extents are not covered by any
 device model.  But, since they are addressable, they can be included in
 redundancy sets.

 This philosophy may lead to ending the current problem with using too many
 device types.  George will be thinking about that.

 Doug preceded to review the lists of devices and commands.  The purpose was
 identifying the correct list of devices.  For each device, Doug sought to list
 the commands that make sense for that particular device.  Also, Doug wanted to
 validate the functions of the various commands.

 11.  Error Handling for SCSI Controllers

 Doug Hagerman presented the first draft of his document describing error
 handling in SDAs.  Doug started by noting that SDAs have much more diverse
 failure modes than just plain disks.

 Doug initiated a discussion of Contingent Allegiance in SDAs.  If a leaf disk
 in an SDA goes into Auto Contingent Allegiance (ACA) due to an operation on
 the array, can a physical path (different) be used to clear the ACA.  Doug
 proposed that the disk views the various path to it in the same way that it
 would view paths from different initiators.  That is, the ACA only can be
 cleared by operations on the path where the CHECK CONDITION status was
 returned.  All other operations on all other paths see an ACA ACTIVE status
 until the ACA is cleared on the proper path.

 Doug proposes no new status codes.  But, George noted that ACA ACTIVE is
 missing for the document's list.  Discussion of the ASC/ASCQ list was more

 George strongly defended the current SCSI method for naming and describing
 ASC/ASCQ assignments.  Doug and several others want to define ASC/ASCQs in the
 SCC document (as device model features).  George prefers defining ASC/ASCQs in
 the SCSI-3 Primary Commands (SPC) and then using them in the SCC.

 Doug next began working through the list of ASC/ASCQ descriptions needed.
 Each listed ASC/ASCQ was discussed in terms of:

  - what it means,
  - does an equivalent ASC/ASCQ already exist,
  - is the proposed description too detailed?

 Significant problems were uncovered for reporting the failing part in the
 REQUEST SENSE data.  The FRU field is only one byte, which is too small by at
 least one byte (and at most seven bytes).  There are no reserved or otherwise
 available fields in the REQUEST SENSE data.  Reformatting the REQUEST SENSE
 data will wreak havoc on the entire SCSI world.  Creating a new command to get
 the data presents at least all the problems already present for REQUEST SENSE.

 The only solution that seemed even slightly credible is some kind of state
 changes log page.  Then a new ASC/ASCQ would signal the host that the log page
 should be read.

 12.  Action Items

  a) Penokie    Revise the SCSI-3 Addressing (93-161) document and present
                this proposal to the next plenary for a vote.
  b) Penokie    Revise the SDA States and Types (93-191) document.  Include
                the states description found in Doug Hagerman's.
  c) Penokie    Revise the SDA Commands and Mode Pages (93-192) document.
  d) Penokie    Revise the SCSI Disk Array Model (93-140) document and present
                this proposal to the next plenary for a vote.
  e) Hagerman   Revise "Error Handling for SCSI Controllers."  Obtain an X3T10
                document number for it.
  f) Hagerman   Revise "Additional Device Models."  Obtain an X3T10 document
                number for it.

 13.  Meeting Schedule

 The next meeting of the RAID Study Group is planned for Feb 15, 1994 at the
 Red Lion Hotel in Austin, Tx.  The meeting is expected to run from
 9:00am-5:00pm.  This meeting will be a joint RAID Advisory Board Host
 Interface Group and X3T9.10 RAID Study Group meeting.

 14.  Adjournment

 The meeting was adjourned at 04:44 pm.  on Tuesday, January 11 1994.

More information about the T10 mailing list