Re: [tcpm] New version of TCP Option for Transparent Middlebox Discovery available

[Melinda Shore <shore at arsc.edu>]

On Sep 24, 2009, at 10:02 AM, Caitlin Bestler wrote:
> The part that makes perfect sense is the desire to have a middle-box  
> discovery probe that follows
> the same path that a TCP packet would.

I think there's a case to be made for in-band
discovery options, which is no something we've
seen previously.  We've had out-of-band path-
coupled and out-of-band path-decoupled, but not
this.  (Next up - in-band path-decoupled ... )

However, I agree about the OUI question, and I
think the stuff discussing problems along a path
could be considered either extremely clear or
extremely muddled, since it doesn't exist.  There
are not just problems around two devices from
the same vendor, but also problems about multiple
types of devices from the same vendor, etc.
Presumably you could infer some stuff about
ordering and topology from TTL data if it were
written into the response.  I'd argue rather
strongly that you'd want an ordered list of as
many of the middleboxes as were discoverable -
this is not that much like tunnel endpoint discovery.

I like the idea of having some sort of in-band
discovery mechanism but I'm not sure about tying
it to TCP (although there's an implied requirement
for a response, which you can't guarantee with, say,
UDP or some other transports) and I'm really not
sure about v4.

Melinda