94-199R1 -- REVISED Minutes of the XOR Study Group Meeting on September 12, 1994

Ralph Weber -- VMS -- ZKO3-4/U14 weber at star.enet.dec.com
Fri Sep 23 15:16:04 PDT 1994


                                                             X3T10/94-199R1

To:       Membership of X3T10

From:     Gerry Houlder
          Seagate

Date:     September 23, 1994

Subject:  Minutes of the XOR Study Group Meeting on September 12, 1994


Attendees:

Gerry Houlder            Seagate
Jay Elrod                Seagate
Paul Hodges              IBM
Bill Hutchinson          HP
Thai Nguyen              Storage Technology  thai_nguyen at stortek.com
James McGrath            Quantum
Edward Fong              Amdahl
Larry Lamers             Adaptec
John Lohmeyer            NCR

Gerry Houlder acted as chairman for the meeting.  The issues discussed are
summarized below.
 
(1) XDWRITE command descriptor block layout - In response to criticism that
the 16 byte CDB didn't follow accepted SCSI guidelines, Gerry Houlder
proposed a slight rearrangement of fields to follow even byte boundaries
and the general structure given in SBC. This proposal included a 3 byte
secondary address field and a LongID bit to allow for an 8 byte address in
the data phase.  Discussion on item (2) subsequently reduced the secondary
address field to one byte and eliminated need for LongID bit.
 
(2) Use mode page to store/define redundancy group addresses.  The group
preferred using a small secondary address field that contains an index into
an internal table of 8 byte addresses.  Gerry Houlder will draft a proposal
for this.  A key problem to be solved is that different areas of a target
may be in different redundancy groups, and different addresses and or
numbers of devices may apply to each group.  A mode page the works like the
notch page (page Ch) will be drafted that allows a definition of redundancy
group ranges as well as the number of group members and address of each
group member.  The XDWRITE, REGENERATE, and REBUILD commands will need a
different (and simpler) structure to make use of the addresses from the
mode page.
 
(3) Transfer of error handling - This was a discussion of document
X3T10/94-184, which was sent on the SCSI reflector earlier.  The group
preferred using 3rd party reservation technique for error recovery (item C
in the document) because it doesn't require defining any new constructs.
The XOR command document will add this procedure.
 
(4) Other error handling issues - The group discussed what the "primary
target" should do if the "secondary target" returns Reservation Conflict or
ACA Active status.  The preferred response was to return Check status to
the initiator and return Command Aborted sense key with new ASC for
Command Blocked.  The initiator should assume that the parity drive has not
been updated and the data drive may be partially updated (i.e., is in an
unpredictable state).  Its error recovery action should include restoring
both the data and parity drives.
 
Action of Busy or Queue Full statuses wasn't discussed, but the primary
target should retry the secondary command a reasonable number of times
before resorting to the Check status w/command blocked response.
 
(5) Multi-controller data validation problem - Paul Hodges (IBM) posed the
problem of one initiator doing an update write (XDWRITE with an XPWRITE to
another drive) while another initiator is doing a regenerate on the same
LBAs.  The regenerate operation could read new data from the data drive
(because XDWRITE is done or a cache hit on new data occurs) and get old
data from the parity drive (because XPWRITE hasn't happened yet).
 
Our conclusion is that this problem is not unique to XOR command
architectures and can only be solved by having RAID controllers co-operate
with each other on such activities.  We didn't identify any particular
implementation rules that should be added to the XOR commands.





More information about the T10 mailing list