XOR Command meeting minutes 11/7/94 -- Document number X3T10/94-247r0

Jay Elrod Jay_Elrod at notes.seagate.com
Mon Nov 21 09:54:39 PST 1994


 
Attendees:
 
Gerry Houlder        Seagate
Jay Elrod            Seagate
Brian Davis Seagate
Jeff Stai  Western Digital I/O Products
John Baudrexl Fujitsu/Intellistor
Colin Schaffer Fujitsu/Intellistor
Bob Snively Sun Microsystems
Paul Hodges          IBM
Stephen Fuld StorageTek
Bill Hutchison      Hewlett Packard
Charles Monia Digital Equipment Corp.
 
Gerry Houlder acted as chairman for the meeting.  The issues discussed are
summarized below.
 
1) Redundancy Group Addressing Mode Page - A description of the new mode page
to do redundancy group mapping was added to the RAID 5 document and was
formally presented at the meeting, along with an alternative more complex mode
page configuration. The more complex version allows each drive in a RAID group
to have a unique logical block address range for that group, rather than a
common range which applies to all drives in the group. Bob Snively (Sun
Microsystems) thought the added flexibility of this configuration may be
desirable, based on the way their (Sun's) RAID controllers are designed. Bob
also recommended that we restrict use of the Address mode page to those
interfaces which require more than 3 bytes for a physical device address (the
primary purpose of the table is to provide a greater-than-3-byte device address
mechanism).
 
2) New mode page effect on operation of XDWRITE, REBUILD, and REGENERATE
commands - the implications of an address mode page as it relates to these
commands were discussed, including the concept of requiring the logical block
address to be a relative address when using the more complex table
configuration described in item 1 above. The various forms of parameters for
the REBUILD and REGENERATE commands were also discussed briefly.
 
3) Possible data corruption problems - Paul Hodges (IBM) presented a scenario
where a Regenerate and XDWRITE command issued to separate data drives in a RAID
group, addressing the same LBA range, could result in improper data
regeneration by the Regenerate command. This would happen if the Regenerate
drive issued a Read to the parity drive prior to the XDWRITE target issuing the
XPWRITE to that same parity drive, but issued the Read to the XDWRITE target
after the XDWRITE was complete. The result is that the Regenerate command would
be using old parity (from the parity drive) and new data (from the data drive).
This would result in bad regenerate data. The concensus was that the controller
needs to ensure that this situation does not happen. This could be accomplished
by  preventing an XDWRITE from being issued any time a Regenerate or Rebuild is
outstanding for an area which overlaps with the extent of the XDWRITE command
(and vice versa). It was pretty much agreed that this is what is already done
in existing RAID controllers to prevent the same problem (stripe locking). It
was agreed that an implementor's note should be added to the RAID document
which advises the controller designer of the need to guard against this
possibility of data corruption. Jay Elrod will add this note.
 
4) 3rd party recovery procedure - It was pointed out prior to the meeting that
the 3rd party reservation procedure agreed to last meeting can still result in
a problem with previously queued commands starting execution before the error
on the parity drive is resolved. Everything from transferring of ACA to not
doing anything was discussed. There is some feeling that it may not be
necessary to do any 3rd party reserves, transferring of ACA, etc. If this is
the case, this issue once again becomes a non-issue. It seems that one question
at hand is "Exactly how much is necessary at the host level when it comes to
error recovery of an XPWRITE?" For the time being, the concensus is that some
sort of REBUILD would probably be appropriate. This still needs thought and
discussion.
 
5) Steve Fuld (STK) pointed out that the NDisk bit in the XDWRITE command may
be redundant, due to the addition of the secondary control field value of 11,
which specifies that the drive save the data in the buffer for the host. This
will be further investigated by Jay Elrod, and the NDisk bit will be removed if
Steve's observation is found to be correct.
 
6) Write caching was discussed, and it was recommended that the RAID document
define at what point status is returned for the XPWRITE command when write
caching is enabled. It was casually agreed that, with write caching enabled,
ending status should probably not be returned for the XPWRITE command until the
xor of new and old data has completed, and the result is in the buffer ready
for writing to the media. Jay Elrod will add this to the document.
 
7) The question was raised by Steve Fuld as to whether there needs to be
entries for both the Max XPWRITE and Max XDWRITE lengths in the RAID Control
mode page. There was no good case made during the meeting for leaving both of
these fields in the mode page. Jay Elrod will investigate this further and
consolidate the two fields into one field if no reason can be found to keep
them both.
--
Jay Elrod -- Jay_Elrod at notes.seagate.com
-------------------------------------------------------------------------
Seagate Technology   -   920 Disc Drive   -   Scotts Valley, CA 95066 USA
Main Phone 408-438-6550   -   Email Problems postmaster at notes.seagate.com
Technical Support: BBS 408-438-8771  Fax 408-438-8137  Voice 408-438-8222  
-------------------------------------------------------------------------

### OGATE Version 8 message trace and attachment information:
### MsgFileName: m:\mgate\outbound\217.MSG
### Org Date:    11-21-94 08:48:44 AM
### From:        Jay Elrod at SEAGATE
### To:          SCSI @ wichitaks.ncr.com @ internet
### Subject:     XOR Command meeting minutes 11/7/94 -- Document number X3T10/94-247r0
### Attachments: none




More information about the T10 mailing list