The Routing Over Large Clouds Mailing List Archive by date

["j.garrett" <jwg@garage.att.com>]

Curtis,

> The statement "For the purpose eeeg loop prevention, it is advisable
> avoid the non-NBMA paths between the routers where NHRP is being run."
> is total nonsense.  What you are stating is that if the Internet
> deploys an ATM network where multiple Internet service providers
> attach, those Internet service providers should never use routes
> learned from other media.  This is an unworkable attempt to whitewash
> the problem.
> 
> NHRP can only safely do address resolution.  You are trying to
> perpetuate the already disproven claim that NHRP is a viable
> replacement for routing.
> 
> Curtis

Amen!

Moreover, NHRP Fabric Mode procedures require that the next-hop
determined by routing be resolved by some means other than NHRP.
That is, with NHRP Fabric Mode, routing IN NOT ALLOWED to find the
"optimal" next-hop unless other procedures are provided to resolve
the next-hop.  The router originating an NHRP request encapsulates
the request in an IP packet addressed to
    "either to the next-hop router or to the ultimate destination being
     resolved."
Thus, the router can't forward the NHRP request UNLESS it
can resolve the next-hop address by some other means.
So if the only reason to send the NHRP request is to resolve
the next-hop address, NHRP Fabric Mode (as written) is useless.

Of course, NHRP Server Mode doesn't scale (well, at all),
and IMHO is also useless.

If NHRP was the only solution available, I would understand
the "better a flawed solution than no solution" argument.
Fortunately, there are at least two different ways to solve the
address resolution problem - (1) embed subnetwork (e.g., ATM) addresses
in the routing protocol or (2) a query/response address resolution
protocol with querys sent to the router that provided the next-hop
information, rather than to the next-hop.  The query/response
approach resolves both next-hop router addresses and destination
host addresses, while embedding subnetwork addresses in the routing
protocol only resolves next-hop router addresses.

Given that we have known for more than a year that NHRP can
create stable routing loops, why must we continue to cling
to NHRP as the only possible solution?

John Garrett