RE: iscsi: rev 05 2.5.1 requires ACA support

[Black_David@emc.com]

> To impart some finality into the discussion, can you summarize for
> the benefit of the mailing list what is distinctive about iSCSI that ACA
> must be made mandatory for targets. I'm not trying to start an argument
> here, but a few words of motivation might help us to understand
> as to why iSCSI-ACA must be made manadatory (required in most
> situations) versus optional (implement where needed).

Actually there are (at least) two issues underlying
Prasenjit's simple question -- this whole area of
target response to errors feels like the old Adventure
"maze of little twisty passages, all different" :-).

Issue 1: Should ACA be REQUIRED, RECOMMENDED, or
	OPTIONAL to implement?
	- Independently for Initiators and Targets.
Issue 2: Should ACA be REQUIRED, RECOMMENDED, or
	OPTIONAL to use?
	- Basically an Initiator issue, as SCSI puts the
		Initiator in charge of deciding whether to
		use ACA.

I'm not sure Julian and I see eye-to-eye on this, so
here are my current opinions:

Issue 1, Initiators - ACA should be OPTIONAL to implement
Issue 1, Targets - ACA should be REQUIRED or RECOMMENDED
	to be implemented (note, this is a slight change from
	my earlier email).
Issue 2 - ACA should OPTIONAL to use, Initiator makes the
	choice in each case.

So, Prasenjit's question to me is "Why should ACA be
	REQUIRED or RECOMMENDED to implement in targets?".

ACA is specified in SAM, and T10 therefore requires that
any transport mapping of SAM specify how ACA is implemented.
Hence it has to be specified for iSCSI, lest we get dinged
by T10 for failing to handle all SCSI features.

This leaves the choice of REQUIRED, RECOMMENDED or OPTIONAL
for target implementation.  My concern is based on looking
forward to the point in time where we work on moving iSCSI
from Proposed Standard to Draft Standard.  At that
juncture, features that aren't implemented will be removed
from the specification (retaining feature X requires at
least two independently developed interoperable
implementations of X).  If ACA is removed at that point,
iSCSI will get dinged by T10 for no longer being a
complete mapping of SCSI, which would be bad.

Hence, the word we choose should lead to multiple
implementations of ACA that can be tested for
interoperability, even if ACA is rarely used in practice.
The removal of features for which interoperability
cannot be demonstrated as part of advancing from
Proposed Standard to Draft Standard is distinctive
to iSCSI by comparison to other SCSI transport mappings.

OPTIONAL strikes me as the wrong choice.  I'd like to
avoid a situation in which iSCSI can't move to Draft
Standard because nobody's implemented an OPTIONAL to
implement feature -- IMHO, that sounds really silly.
REQUIRED seems most likely to lead to implementations.
RECOMMENDED would also be ok.

Sorry to have to give such a long answer to a short
question,

--David

---------------------------------------------------
David L. Black, Senior Technologist
EMC Corporation, 42 South St., Hopkinton, MA  01748
+1 (508) 435-1000 x75140     FAX: +1 (508) 497-8500
black_david@emc.com       Mobile: +1 (978) 394-7754
---------------------------------------------------