Re: [RAM] Comment on draft-farinacci-lisp-00.txt (LISP)

[Dino Farinacci <dino at cisco.com>]

> Dino just mentioned how LISP 1 is a "phase 0 prototype effort", and  
> I suspect
> the same is true, in some ways, of many of the lower numbered LISP  
> variants.
> My sense is that the one that gets deployed in large-scale operational
> service is likely to be LISP 4 or 5 (or more), i.e. one that's not yet
> defined.

But please note, that these are variants and not version numbers.  
That is, each number is using a particular method to do mapping. So  
LISP 1.x was suppose to convey routeable IDs, LISP 2 was suppose to  
convey a pull model and 2.0 is using DNS. And 3.0 is future stuff.

So for instance (and this is not real yet and hasn't been written  
down), this is how I would number the variants:

LISP 1 and 1.5: as documented in draft-00.

LISP 2.x:       uses a pull model where:
    LISP 2.0:   uses DNS with port 53
    LISP 2.1:   uses DNS the protocol on a different port and  
infrastructure
    LISP 2.5:   uses a new and different pull protocol

LISP 3.x:       uses a pull model that does not exist today, where:
    LISP 3.0:   uses DHTs
    LISP 3.1:   considers using Compact (or ROFL) Routing

LISP 4.x:       uses a push model that does not exist today, where:
    LISP 4.0:   uses BGP not on port 179
    LISP 4.1:   uses Mark Handley's DNSpush style
    LISP 4.2:   uses a push-n-pull model, where push happens at high  
levels and
                lower levels pull from the higher levels

LISP >= 2 means:

o Never routable IDs.
o Could mean packets are dropped while waiting for lookups to complete.

All variants of LISP assume:

o The mapping function does not convey locator reachability status.
o The EID-to-RLOC entries are relatively static, that is they change  
only at
  subscription time events.

> My guess is that the architectural commonality between LISP 1/1.5/ 
> etc and the
> eventual deployed stuff is likely to be:
>
> - Hosts and local routers don't need to be modified

ISP routers that don't use LISP for TE won't have to be modified either.

> - The existing internetwork layer is "jacked up" to become mostly an
> 	end-end host naming layer
> - End-end names are mapped into new locators as they cross the  
> boundary

Right.

> I think LISP is the first detailed proposal in the last half-decade  
> or so to
> propose operating in this particular architectural quadrant (which  
> may well be
> the only feasible one to operate in), and I suspect that's why it's  
> getting a
> lot of attention. However, the final product may look quite  
> different from the
> initial prototypes we have on paper now.

Definitely. We are trying to draw a line in the sand to start off,  
but the line isn't really that deep.  ;-)

Dino

_______________________________________________
RAM mailing list
RAM at iab.org
https://www1.ietf.org/mailman/listinfo/ram