RE: iSCSI : Command Ordering Proposal.

[Charles Monia <cmonia@NishanSystems.com>]

Hi Santosh:

See comments below.

> -----Original Message-----
> From: Santosh Rao [mailto:santoshr@cup.hp.com]
> Sent: Friday, January 26, 2001 3:17 PM
> To: Charles Monia
> Cc: IPS Reflector
> Subject: Re: iSCSI : Command Ordering Proposal.
> 
> 
> Charles Monia wrote:
> 
> >   For example, if I
> > send commands A B |C| D E, where |C| is ordered, I expect 
> that D and E won't
> > complete before |C|.  If A B D and E arrive followed by 
> |C|, there's no way
> > to obtain the correct behavior. Incidentally, similar 
> constraints hold true
> > for "head of queue" tasks.
> >
> 
> Charles,
> 
> SAM-2 Section 7 states :
> 
> "The rules for task set management apply only after the task 
> is entered
> into the task set."
> 
> This implies that the enforcement of the ordering of task 
> tags in your above
> example is NOT based on when the initiator sent the commands, 
> but on when
> these tasks enter the task set at the target.
> 
> i.e. The initiator may send A, B, |C|, D, E, where |C| is ordered.
> 
> The commands may arrive at the target in the order :
> A, E, |C|, B, D in a multi-connection session.
> 
> The commands A and E are not subject to any form of ordering.
> (being simple tag commands and having no ordered tag commands
> ahead in the task set.)
> 
> The commands |C|, B & D are subject to the ordering that B & D cannot
> be executed until C is first executed. Thereafter, B & D are 
> subject to
> simple tag rules.
> 
> I believe the confusion lies in whether ordered task tags 
> imply end-to-end
> ordering or ordering within the received task set at the 
> target. the former
> requires strict ordering on the link or re-ordering at the 
> target, in the
> absence
> of strict link ordering. The latter does NOT require any link 
> ordering or
> re-ordering
> at the target, since it only enforces ordering within the 
> received task set.
> 
> Comments ?

Aside from the fact that the ordering semantics make no sense without this
property, the initiator's real-world expectation is that end-to-end ordering
is preserved. Also, as I mentioned earlier, without end-to-end ordering,
there is a potential for other undesirable side effects due to the
misordering of task management requests relative to commands in flight.

Charles