Re: router-router NHRP

[[e-mail]]

The recent proposals of Yakov/Bruce for router-router NHRP loop
suppression are very interesting, and (alternatives 1 and 2) bring out
some important facts, namely, that state information needs to be
tracked along the routed path between NHRP query sender and responder.
As is mentioned in their e-mail, there is a scalability problem in the
first 2 alternatives, because of the need to maintain state and take
action proportional to the the number of NHRP requests.  Alternative 3
is based on loop detection, does not have the aforementioned scaling
problem, but requires loop detection algorithms capable of rapidly
detecting loops under many varied conditions to be defined.

Wrt the scalability problem in alternatives 1 and 2, the scalability
problem seems to be related to the fact that allowing cut-though via a
local caching mechanism has the affect of dis-aggregating the routing
information that typically would have been aggregrated in a large
Internet. The forwarding of the NHRP query via the router
infrastructure is feasible in a large Internet, because of the
scalability afforded by routing aggregation procedures. Once the
cut-through is accomplished, the scalability afforded by routing
aggregation seems to go away, because there is a direct association
between source and destination. This seems to suggest that loop
prevention techniques along the lines of alternatives 1 and 2 may be
effective only a smaller scale (e.g., when the number of routers in
the NBMA is a few 100s). This would also suggest that if alternative 3
could be shown to work, it would be preferred.