Re: [RRG] Question about ITR and ETR deployment

[Lixia Zhang <lixia@CS.UCLA.EDU>]

On Apr 1, 2008, at 8:53 AM, Scott Brim wrote:
> Excerpts from Lixia Zhang at 22:56:42 -0700 on Mon 31 Mar 2008:
> > On Mar 31, 2008, at 7:51 PM, Gang Chen wrote:
> > > Dear Tony and Lixia:
> > >       May I ask you question about ITR and ETR deployment location?
> > > In the LISP draft-farinacci-lisp-06.txt, both CE and PE router can
> > > be used  for tunnel endpionts. I wonder to know where is the optimal
> > > location to reduce router table size in DFZ greatly?
>
> > *personal* view:
> > - the exact location of tunnel end point (PE or CE), I believe, does
> > have an impact on the details of a design, and more importantly, on
> > the incremental rollout (if interested, you may want to take a look  
> > of
> > the APT incremental deployment presentation slides from Philly  
> > meeting)
>
> The more we take these ideas and try to apply them to real life
> operations, the more we realize how helpful coordination among xTRs
> is.  In order for a site to be able to coordinate behavior of its
> xTRs, to any degree and in any way it chooses, it is good to have them
> under the site's direct control.

I completely agree with the above.

(although I suspect, maybe only slightly, that what we have in mind  
about "site" in the above statement could be different, i.e. a  
customer site vs an ISP AS)

>  I don't believe we can predict or
> should control how a site's xTRs will need to work together -- they
> might do something surprising.  So I don't think a site should depend
> on inter-provider cooperation.

again I strongly agree that one AS should not depend on others  
cooperation.

and following all the above I had expected the next sentence to say  
that it's better to place xTRs in PE, but it said the opposite! :-)

>  That means I believe that it's better
> to place xTR functionality in CEs.


this makes a perfect example to illustrate my concern in the earlier  
reply Robin: even when people may make the same statement, that does  
not necessarily mean we all understand each other.  So gaining a real  
understanding, rather than critique based on incomplete understand, is  
a first important thing.

Scott and I had a skype chat earlier which helped me solve the above  
puzzle. Below is my summary of that conversation; Scott can help  
correct for any part I got wrong.

To deploy encapsulation requires two points coordination, the encap  
point (ITR) and decap point (ETR) (I tried to avoid the word  
"tunneling" here, as Scott pointed out that tunneling is often used to  
refer to configured tunnels)

Scott considered CE the right place, based on the assumption that CE  
can decides whether to perform encap, based on its understanding of  
whether the other end know the new game to do decap.

Why I concluded PE as the right place: since we said earlier that
(1)to deploy encapsulation requires two points to collaborate, and
(2)one AS should not depend on others cooperation to deploy  
encapsulation.
Furthermore, whoever making the first deployment needs to have an  
incentive (some payback)
Putting all the above together: if an ISP is a first mover:
- it can have both ITR and ETR be "under the site's direct control",
  i.e. they'd be the ISP's border routers (PEs), so traffic get  
tunneled
  through the ISP
- as I explained in Philly's talk, there is a good hope that the ISP can
  get payback in terms of reduced BGP table size.

Lixia

--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg