draft-dube-route-reflection-harmful-00.txt

[Tony Li <[e-mail]>]

Rohit, John,

Thank you very much for your recent draft.  I believe that it points out
two important points that are not dealt with adequately in RFC 1966.

As you note, there are route reflector configurations that are not tenable.
This should have been noted.  However, your avoidance guidelines for this
problem, if I understand them correctly, are overly restrictive.  In
particular, in your example, you can see that the IBGP session between
route reflectors is safe if the membership of R3 and R4 is reversed.  More
generally, the restriction is that clusters need to be concave: that is,
the IGP must never compute a path that exits a cluster and then returns to
that same cluster.

Also, it probably should have been emphasized that route reflectors do not
compute the same results as a full IBGP mesh.  I believe that this is and
was widely known, and that there are many other opportunities to observe
this in addition to scenarios similar to the one described in your draft.
This should certainly be fully documented, if only for truth in
advertising.

My final comment is about the title of this draft.  While I'm well aware of
the history of the genre that sparked this title, one of the implications
of the title is the surface impression that you are trying to discourage
others from employing route reflectors.  I don't believe that you actually
hold that viewpoint, given the contents of section 7, so I think that this
choice of title was somewhat unfortunate.

I believe that an appropriate way of addressing this comment would be a
update of RFC 1966.

Regards,
Tony