Don't Change QErr Definition

PJohansson at aol.com PJohansson at aol.com
Fri Nov 22 12:27:44 PST 1996


* From the SCSI Reflector (scsi at symbios.com), posted by:
* PJohansson at aol.com
*
In a message dated 96-11-21 12:55:14 EST, ROWEBER at acm.org writes:

<< Problem
 
 In a multi-initiator environment, the current definition of QErr=1 can 
 produce a situation where no work is getting done because the device 
 server is continually reporting CHECK CONDITION status.  Consider the
 following scenario:
 
 	Initiator A		Initiator B
 	Send Tag 1
 	Send Tag 2
 				Send Tag 3
 	CHECK CONDITION for Tag 1
 	>>Tags 2 and 3 are aborted<<
 	Send REQUEST SENSE
 	Send Tag 4
 				Send Tag 5
 				CHECK CONDITION for Tag 5
 	>>Tag 4 is aborted.<<
 				Send REQUEST SENSE
 				(UNIT ATTENTION for aborted tasks)
 				Send Tag 6
 	Send Tag 7
 	CHECK CONDITION for Tag 7
 	>>Tag 6 is aborted.<<
 	Send REQUEST SENSE
 	(UNIT ATTENTION for aborted tasks)
 
 I could go on and on, but you get the picture.  If the hosts keep the
 device server busy enough, there is a chance that the only work that 
 will ever get done (after the first error is encountered) is moving UNIT
 ATTENTION sense data around.
  >>

Besides agreeing with the other comments against Symbios' proposal to alter
the definition of QErr, I wonder if this is a real-world situation that has
been encountered and is causing problems.

I would not be at all surprised if initiator code that knows it is working in
a multi-initiator inserts some modest delay before sending any new commands
after observing an additional sense code of COMMANDS CLEARED BY ANOTHER
INITIATOR. This permits a form of hysteresis and prevents the sort of cascade
sequence that Ralph describes.

In short, is the cascade scenario an artificial example?

Peter Johansson

Congruent Software, Inc.
3998 Whittle Avenue
Oakland, CA  94602

(510) 531-5472
(510) 531-2942 FAX

pjohansson at aol.com

PS Gerry, I was glad you said, "Just for the record, I don't like this
change" at the end of your message. Up to that point I was having trouble
determining what you thought of the proposal! :^)
*
* 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