RE: iSCSI: Out Of Sequence due to null sequence with multiple connections.

["Binford, Charles" <CBinford@pirus.com>]

See comments below, marked with ++++

Charles Binford
Pirus Networks
316.315.0382 x222


-----Original Message-----
From: Black_David@emc.com [mailto:Black_David@emc.com]
Sent: Monday, April 02, 2001 2:13 PM
To: CBinford@pirus.com; ips@ece.cmu.edu
Subject: RE: iSCSI: Out Of Sequence due to null sequence with
multipleconn ections.


It still appears to me that the behavior that
Charles wants is available by sending the task management
function twice - once for immediate delivery and once for
ordered delivery.  Duplication on all connections would
not be necessary in the normal case because the ordered
instance would naturally come behind the unordered ones.
The timer seems to be something that ought to be taken
up under error recovery in general (and in particular,
we should consider letting TCP time out the connection
rather than adding yet another recovery timer).  IMHO,
the bottom line is that the logic to duplicate the task
management commands at the Initiator and coordinate
them at the Target costs implementation complexity,
and I have not seen a convincing statement of what
we're buying with that added complexity.

It may be the case that applications don't want to be
bothered with sending task management commands twice,
in which case we have a coordination problem of some
form that has to be addressed in iSCSI.

+++++++++++++++
I would state this much stronger.  Applications had better not have to know
that it is iSCSI underneath vs. FCP or parallel SCSI else I believe we
missed the objective (granted, some things such as target address space are
unavoidably different, but I believe task management functions should be the
same).  The transport needs to handle the transport issues without exposing
quirks to the SCSI or application layer.

I would be more agreeable if David was advocating the iSCSI layer was
responsible for sending task management functions twice, once with, once
without ordering.  This puts the handling of the iSCSI specific ordering
issue in realm of iSCSI.  My caution to this approach is that is changes the
behavior of the target slightly.  The SCSI target layer will see the task
management function twice, not once.  I can't think of any scenario where
this would matter, but I get nervous when we change behavior from the norm.
++++++++++++++++++ cb


For further discussion,
--David