Re: Comments on NHRP spec.,, modes of deployment etc.

[Bruce Cole <[e-mail]>]

> I understood the original question to request a clarification
> of the fact that there can be two identifiable 'interfaces' involved
> in NHRP (the host-NHS and the NHS-NHS), even though the same
> packet format is used to propagate NHRP messages in each case.
> So, in one sense there is indeed a server to server 'protocol',
> it just happens to be the same as the client-server case.

I don't see a distinguishable difference between the two.  'hosts' still need
to be able to send NHRP responses in addition to NHRP requests.  (The ability
of 'hosts' to answer NHRP requests is still required since it was not
made a requirement that NHSes assume the full burden of providing
NHRP responses for stations that attempt to register with them).


> The alternative interpretation is that every NHS is considered
> a client of the NHS it forwards the query to, so ensuring
> you only ever have a client-NHS case. Is this they way you
> visualise it? (forgive me it says this in the spec - my memory
> is not what it might be sometimes :)

I visualize every NHRP station behaving as a client, and some stations
additionally behaving as forwarding agents and possibly providing
non-authoritative NHRP responses.

Ignoring router-to-router support, it would seem to me that the only
operations that a 'host' need not perform are:
    providing NHRP responses for stations other than self
    forwarding of NHRP packets

And David writes: 

>What I had in mind was that the NHRP clients register with, send queries to 
>and get responses from NHRP servers. NHRP servers forward/resolve queries, 
>and return responses to clients. These are different functions.

Except that the clients resolve queries, and send responses too.  And
the servers probably register with their next hops too.  So the functions
become less distinct.

>Listing the set of operations that an NHRP client must perform,
>and then listing the operations that a server must perform would make
>it clearer to potential implementators what needs to be built for a client
>versus a server.

I would think that it is 'obvious' that clients need not deal with forwarding
of NHRP packets.  Besides that, I'm not sure what kind of description
would be appropriate.  Maybe some proposed text would help (he suggests
lazily :-)).