[arin-ppml] Routing Research Group is about to decide its scalable routing recommendation
I found these presentations interesting:
concerning problems with iBGP in large ISPs.
> I find it interesting that folks in RRG are worrying most about
> the number of unique *prefixes* in the DFZ, while the number of
> unique *routes* (prefix/path attributes) is growing steeper.
I understand "unique route" means a best path received from a
neighbouring router. Each router chooses one of these to be the best
path it uses for its FIB, and which it offers to its neighbours,
according to local policy and various algorithms.
> Further, most all the work in the IETF BGP-related working groups
> aims to ADD more routes in order to support multi-path, alleviate
> route oscillation, provide faster convergence, obtain more operations
> visibility, etc.., pretty much completely in odds with the RRG work,
> and exacerbating the issues of scalability and instability, path
> hunting, RIB issues, RAM, CPU, FIB I/O, etc..
I don't know much about how changes to BGP will cause more work for
routers, but I can easily imagine this being the case.
> Perhaps the disparity between the IETF work and the IRTF RRG work
> is that the operators are busy running networks and dealing with
> real operational issues, while the RRG folks are solving problems
> that won't been an issue for a decade or more.. This seems eerily
> similar to the IPv6 conundrum as of late, eh?
Maybe so. However, as long as the RRG work does result in a
reduction in the number of DFZ routes, below what would have happened
without this work, then this will be worthwhile. Maybe it will be
another 10 years or more before it happens - but it will still be
well worthwhile, I think.
I think that the proposals which could possibly be voluntarily
adopted on a wide enough scale to solve the routing scaling problem:
are core-edge elimination schemes. LISP-ALT and Ivip so far have
been submitted for the final RRG process. I assume APT and TRRP will
be submitted too.
All these are intended to provide a much greater number of end-user
networks having their own scalable PI space (my term) in a way that
each such prefix doesn't need to be in the DFZ. Since the growth in
"unique routes" you refer to is due to a growing multiplier effect on
the number of DFZ routes, one of these schemes being successfully and
widely adopted should offer a big relief.
The other proposals, I think, are core-edge elimination schemes. I
can't see how these could be voluntarily adopted, since they require
major host changes (stack and apps). They too would provide scalable
multihoming, portability etc. without burdening the DFZ with a prefix
for each such end-user network.