RE: iSCSI: No-renegotiation rule inadequately described

[Martins Krikis <mkrikis@yahoo.com>]

--- pat_thaler@agilent.com wrote:

> The problem is there are two ways of
> interpreting the state of the negotiation 
> after ImmediateData=Reject. 
> 
> A) The offered value was rejected so the
> negotiation isn't done yet. In this view
> the second offer is a continuation of the
> first negotiation rather than a new 
> negotation.

Good point. But I don't think this view quite
fits together with, e.g., page 69:

  Reject or Irrelevant are legitimate
  negotiation options...

In my opinion (and implementation) it also
complicates the decision about whether this
value is precluding me from announcing readiness
to commit or it isn't. It's harder to view it
negotiated for commit purposes and non-negotiated
for repeated reception purposes than it is to
just view it the same way...

> B) A negotiation is always one offer and one
> response regardless of whether the response
> indicates successful negotiation or not.
> 
> Text should be added to the draft to indicate
> which is intended. 
> 
> I agree that it should state which is intended.
> My preference would be that it take the simpler
> view in B. 

I'm very much for it.

> For Booleans and numerical values,
> if the first offer caused a reject,
> there is something broken. For lists the originator
> can include all acceptable responses in the initial 
> offer. In either case, there isn't much point in 
> offering again.

This is exactly what I was planning to say once
somebody gave me a reason about the necessity
to renegotiate.

> If the response was NotUnderstood
> then there isn't any point in offering the key again
> (though I also think it is unreasonable to expect
> the
> responder to keep a list of NotUnderstood keys so
> that
> it can detect attempts to renegotiate).

Exactly.

> Irrelevant indicates that a problem such as key 
> negotiation being done in the wrong order (e.g. 
> attempting to negotiation IFMarkInt before
> IFMarker).
> There is no reason this should happen so we don't 
> need to allow the negotiation to continue in this
> case either.

First, there are no dependencies (even for markers,
keys can be viewed as independent), so in my opinion
this cannot happen. But even if it could happen, the
earlier-negotiated key that caused the 
later-negotiated key to be answered with Irrelevant,
is not going away, so there is no point repeating
the later key---it will still be answered with
Irrelevant! There is also no point to negotiate
it from the other side, as that would then be the
side who first called it Irrelevant. Once irrelevant
is always (through the life of negotiation sequence)
irrelevant, so no need to renegotiate it.

> I suggest adding to 4.2 after "Reject or Irrelevant
> are legitimate...." 
> "A negotiation is considered complete when the
> responder has sent the key value pair even if the
> value 
> is "Reject", "Irrelevant", or "NotUnderstood".
> Sending 
> the key again would be a re-negotiation."

This would be truly beatiful. Can it please happen?

Thanks,

  Martins Krikis, Intel Corp.

Disclaimer: these opinions are mine and may not
            be those of my employer



__________________________________________________
Do You Yahoo!?
LAUNCH - Your Yahoo! Music Experience
http://launch.yahoo.com