[c-nsp] Injecting Routes Remotely

[Crist Clark]

We are trying to come up with a way to inject routes into an IGP
remotely. The problem is that there is a device in the midst of the
network which does not speak any routing protocols, which really doesn't
do routing, but is really a bridge for that matter. Here are some more
details of the simple setup:

                        [ leaf nodes ]
                            __|__
                           |     |
                           | RAS |
                           |_____|
                              |
                            __|__
                           |     |
                           |  C  |
                           |_____|
                              |
                         [ network ]

The RAS is the dumb device. We are stuck with this device. We can make
no changes to this device. It usually assigns a IP addresses to its leaf
IP nodes from a known IP network pool (via PPP). So router C (a Cisco
router), just has a static route to this network and distributes this
route into EIGRP. We actually have several of these setups. That is,
"[ network ]" has several of these nets hanging off of it. Each RAS has
its own unique range, so each C router advertises that range into EIGRP,
and all is good.

Now, the problem is that special leaf nodes will be getting IP addresses
that are not within the uinque range to each RAS. Each special node
will get a fixed IP address no matter which RAS it attaches to. We need
to somehow get these routes into EIGRP or else things don't work for
the fixed-IP hosts.

As I stated, the RAS is not up to the task. Instead, there is an server
external to all of this that will know when and where a special leaf node
connects and that node's fixed-IP address. Now, how do I get the information
off of this remote server, and inject it into the routing table? I guess
I should rephrase that. There are many, many unthinkably kludged up ways
to do that, so what I am really asking is, what is the most simple, cleanest,
with as little administrative overhead as possible way to inject these
routes into the EIGRP? Some sub-optimal thoughts we've already had are
to do a scripted interactive CLI session to C to add and delete static
routes. Or insert an extra host, like a *nix system, on the network
between the RAS and router C that can talk to the remote server with leaf
node information and then inject routes to C using RIPv2 or the like. Or
a variation of that where instead of physically putting a host between
each RAS and C, we have a few with GRE tunnels to C to pass RIP through.
We've also have had evil thoughts about using BGP, since BGP peers need
not be adjacent, which is our whole problem here, but haven't developed
those past the "what if we tried..." stage.

I'm hoping someone else out there has had to deal with dumb devices
like our RAS or for some other reason, inject routes into EIGRP from
a remote host that is not a router. I also thought network engineers
might find this a fun little challenge. We did until we saw how painful
it was going to be to maintain all of the nifty hacks we thought up.
I'm hoping we're missing some glaringly obvious way to do this easily,
but I'm not counting on it. Thanks for any help.
-- 
Crist J. Clark                               crist.clark at globalstar.com
Globalstar Communications                                (408) 933-4387