Re: [grobj] Referral definition and its purpose?

[Scott Brim <[e-mail]>]

Excerpts from Keith Moore on Thu, May 27, 2010 06:48:30AM -0400:
> On 5/27/10 3:29 AM, bo zhou wrote:
> >
> >
> >     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.
> >
> >  
> > [bo] I am not sure why B will use P1 to connect to A, because it is
> > hard for B to understand which path is the best. A send B a referral
> > over P1 is bacause only P1 is known at the beginning of
> > communication. A can tell B there are three paths, but cannot tell B
> > which one is the best.
> It is indeed difficult for B to know which path is best.  And in many
> cases "best" depends on the specific needs of the application.   One
> application might need maximum bandwidth, another low delay, another
> might prefer a stable address.  An IPv4 only host would want to use an
> IPv4 address for its peer.  etc.
> 
> Last I knew there was a current IETF WG looking at a similar problem -
> though its name escapes me at the moment and I haven't been following
> their work.

IETF/IRTF work that I can think of that is relevant: NSIS, EME
research group, the MPLS one about using management entities to set up
LSPs, Shim6, MPTCP, LISP.  I'm sure there are more.  But those are all
about mechanisms for determining good paths.  GROBJ is about how to
represent referrals.  GROBJ shouldn't concern itself with how paths
are chosen, just with how to represent those results.