Request/Response Ordering

PJohansson at aol.com PJohansson at aol.com
Wed May 1 08:52:42 PDT 1996


* From the SCSI Reflector, posted by:
* PJohansson at aol.com
*
To the (disorderly?) crowd discussing order:

I think the real zinger can be found in SAM in section 4.6.2.  Check out the
last paragraph.

Alas, SAM makes an assumption that the service delivery subsystem has
in-order delivery as one of its properties but it makes NO REQUIREMENT.  If
this were a requirement, most of our problems might be resolved IF the
operating systems vendors and peripheral vendors would write their
firmware/software to comply.

For example, with Serial Bus it is possible to guarantee delivery order of
request packets to a particular destination if the requester originates no
new request to the same destination until a response has been received for
the first.  In the realm of command packets, that carry CDB's, this does not
necessarily impose a performance penalty if the commands are organized as
linked lists in system memory (where order is explicitly preserved) and the
starting address of the list is delivered to the target.

On the response side, matters are a little more complicated but ordering
requirements are still achievable.  During the course of 'normal' operations,
when the initiator does not send task management functions (whose responses
have order relationships with the responses of the affected tasks), the order
of the return of status for SIMPLE tasks is immaterial, since the tasks may
have been reordered.  In the exceptional cases, where there is an error
condition or a task is aborted or the task set is aborted or cleared, the
target is capable of insuring in-order delivery of the responses by adhering
to the strategy described above --- don't send a subsequent response to the
same node until the (Serial Bus) response has been received for the first
response packet.

Bill is quite right to assert that these problems of ordering and ACA have
received insufficient attention outside of some of these standards committees
--- but that is our own fault.  We need to educate more, to evangelize more
AND make sure that the standards documents themselves indicate just how
critical the issues are.

Peter Johansson

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

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

pjohansson at aol.com





More information about the T10 mailing list