Re: [grobj] Referral definition and its purpose?

[Keith Moore <[e-mail]>]

On 5/27/10 12:31 AM, Dan Wing wrote:
>> 	The definition of referral should be tightened up slightly
>> 	that the parties are exchanging the information via one
>> 	channel (e.g., SIP signaling) but want to communicate (as)
>> 	directly (as possible) with each other *not* using that 
>> 	channel. 
>>
>> that's certainly one case.  another case is when parties A 
>> and B exchange information about how either A or B can 
>> contact one or more other parties.
>
> Two party case is easier to understand and conceptualize.
> This covers a *huge* number of important cases.
agree.
> Then, in a later section number of the same document, we
> can explain 3 party case.
I think I'd rather have the various examples together in the document,
but I won't try to nail down the document outline just yet.
>> also, the referred-to path doesn't have to be more direct 
>> than the path over which the referral is taking place. 
>
> Sure.
>
> But my point is that it is a *different* path than the
> signaling path.   I agree the path could be longer and 
> could go around the moon - twice.
it's _probably_ a different path because if it were the same path,
there'd be little need for the referral.  just reuse the same source
address, destination address, destination port as seen by the initiator
of the connection (though that might not work if the new connection
needed to be initiated by the receiver of the original connection).

but to be general, it shouldn't _have_ to be a different path.  it
should be possible to use a referral that specifies the same path used
by the connection used to send the referral.   (or to put it another
way, do you really want the API to say "sorry, you can't use this
referral because you received it over the very path that it specifies?")

here's an example: over path P1, A sends to B a referral that specifies
three paths by which A can be reached: P1, P2, and P3.  when B tries to
reconnect to A, its circumstances are such that P1 appears to be the
best of the three paths.  I don't know why B shouldn't use P1.
> Do you have a better way to represent that in ASCII that
> gets the point across?  Picture worth a thousand words
> and all..
I thought your picture was okay for the case you were describing, but I
probably didn't look at it too closely as this is all familiar to me
already.

Keith