Re: iSCSI: comments to iSCSI rev 6

[Stephen Bailey <steph@cs.uchicago.edu>]

All,

I agree with Matt.

Every text key in the iSCSI specification should have a precisely
specified response to indicate that it was received and understood.
This does not include X-* keys, for which you're on your own.

A property which I think Matt has articulated, but I want to repeat,
is that an implementation should be able to sanity-check all
negotiation interactions as well as simply communicating the relevant
information.  Specifically, if a negotiation `fails', an
implementation must distinguish between incompatible capabilities and
logical errors whenever possible.

For example if a target sends an AccessID=xxx, the present
specification seems to suggest that there is no response required or
allowed for that key.  I believe we should specify that some response
(perhaps echoing it?) to distinguish between the cases where the
AccessID had the desired effect, and where it had no effect at all
(was ignored).

Explicitly specifying responses for each key allows a more `anal'
implementation to sanity check a more lax one.  We all know we'll
certainly see both implementation styles regardless of what's
specified.

Steph