[c-nsp] IS-IS Multiarea on 12.2 SR

[Jared Gillis]

Oliver Boehmer (oboehmer) wrote:
> Jared,
> 
> Well, don't really know. It's not tested, but it might work in some
> environment/releases.. never looked at it really..

Here's a quick lab diagram/config snippet for anyone who's interested:

A----B
|\  /|
| \/ |
|/  \|
C    D

All routers are 2620XM running 12.3 ipservices latest.

A and B are multiarea L1/L2 routers:

Router A:
int Fast0/0
desc To C 
ip address 192.168.0.1 255.255.255.252
ip router isis C

int Fast0/0
desc To D
ip address 192.168.0.5 255.255.255.252
ip router isis D

int Ser0/0
desc To B
ip address 192.168.255.1 255.255.255.252
ip router isis

router isis
net 00.000c.30ca.5c00.00
is-type level-2-only

router isis C
net 00.000c.30ca.5c00.00
is-type level-1

router isis D
net 00.000c.30ca.5c00.00
is-type level-1

B is similar, with different IP/NET addresses.

Router C:
int loopback 1
ip address 10.0.0.1 255.255.255.255

int Fast0/0
desc To A
ip address 192.168.0.2 255.255.255.252
ip router isis

int Fast0/0
desc To B
ip address 192.168.1.2 255.255.255.252
ip router isis

router isis
 net 00.000a.f49d.9640.00
 passive-interface loopback 1
 is-type level-1

Router D is similar.

In this configuration, routers A and B learn all routes in the network, and exchange them via their L2 link.
Routers C and D are only aware of their directly connected routes, plus a default towards A/B. C does not have Ds routes, and vice-versa, however they are able to ping each other's loops, by following default to A/B which do have the route towards the loop.
I have also taken down the mesh-style connection between A/D and B/C, so the network looks like:
C---A---B---D
And the design works exactly the same.
When I replace A with a 7204VXR running 12.2 SR ipservices, the whole thing breaks. C has no default towards A, and B does not learn any routes that C advertises to A.
The design constraint I have is that in my production network, the C/D routers will be 3750s, which do not have the TCAM space to learn every route in the network I am building, and they will always be a stub (or more exactly an OSPF TS-NSSA), so that's the behavior I am looking for.
I could move to OSPF, but this network will utilize MPLS, and I want to use the MPLS TE extensions of IS-IS. I am aware that OSPF has similar extensions, but IS-IS works better for us, and the network is already built on IS-IS, and an IGP migration is something I'd like to avoid if possible.


> Hmm, if you stick all L1s into the same area (i.e. "standard" design),
> you can't prevent them from seeing the L1 LSPs from the other L1s in the
> area. However you could investigate filtering the routes from being
> entered into the RIB, similar to "distribute-list in" command in OSPF,
> which doesn't exist in IS-IS. But you could try
> 
> router isis
>  distance 255 ip 
>  distance 115 0.0.0.0 255.255.255.255 10
> !
> access-list 10 permit 0.0.0.0
> 
> to have only the default-route in the RIB. Not sure if this helps, not
> sure which problem you are trying to solve :)

That is interesting, I shall have to play with it. As I noted above, I'm trying to emulate an OSPF TS-NSSA in IS-IS, because my stub area routers don't have the TCAM to handle every route in the domain. I just have trouble believing that in 2009 a widely-used routing protocol like IS-IS doesn't have some way of handling this case.

> 	oli