No subject

[GFRAZIER at ausvm6.vnet.ibm.com]

I recently posted a justification for retaining the requirement that targets
terminate ambiguous exchanges after an Abort Task Set, Clear Task Set,
and various forms of logout. However Kurt Chan pointed out that in certain
restricted configurations, removing this requirement would cause no
harm. (I included his note below for reference.) I believe that the present
requirement should still be retained, however.

While there exist certain restricted environments which can tolerate
targets which do not terminate ambiguous exchanges, there also exist
environments in which data integrity problems result when targets do
not terminate them. Therefore, if FCP does not continue to require all
targets to terminate ambiguous exchanges, then initiators will have to
vary their error recovery procedures depending on the exact FCS
configuration, fabric parameters, and N_port parameters. This causes
extensive transport-unique programming, which is one thing the
layering of SCSI-3 is trying to avoid.

                                          Giles Frazier
                                          IBM Austin
                                          gfrazier at ausvm6.vnet.ibm.com

**************************REFERENCED NOTES*******************

 | I do not think this requirement should be eliminated.
 | There are many reasons FCP requires ambiguous exchanges to be aborted,
 | but one of them is in the following case:
 |
 | 1. The initiator has given sequence initiative to the target and the
 | target has been sending data frames to the initiator and some of the
 | frames are unacknowledged.
 | 2. The Initiator sends a Clear Task Set.
 | 3. If the target does not abort the exchange, then the initiator may
 | assume that it can reuse the exchange and sequence ID in a new
 | exchange.
 | 4. Later, the frames from the ambiguous sequence may reappear and
 | cause a data integrity problem.
 |
 | For this case and other reasons, I do not believe that FCP should be
 | changed. Targets must continue to be required to terminate exchanges
 | in ambiguous states.
 |
 |                                  Giles Frazier
 |                                  IBM Austin

 There are two valid topologies where (4) above will never happen:

 - closed loops
 - fabrics which are enforcing sequential delivery in conjunction
   with Targets which will not retry frames on BSY.

 The FCP requirement is therefore unnecessary for a broad range of
 cost-reduced systems.

 Regards,

    Kurt Chan               Hewlett-Packard        System Interconnect Lab
    kc at core.rose.hp.com     Voice: 916-785-5621    Fax: 916-785-2875
    8000 Foothills Blvd     MS R5NF                Roseville, CA 95747
****************************************************************

               (Please see my response above.)
               Giles Frazier
               IBM Austin
               gfrazier at ausvm6.vnet.ibm.com