[c-nsp] OSPF router gets separated from a broadcast domain

[Peter Rathlev]

Hi Gabor,

I can't see how you could avoid this problem. If you have the following:

          Host1    Host2
            v        v
RTR-A <-> SW-A <-> SW-B <-> RTR-B

and you break the link between SW-A and SW-B, what would you suggest
would determine if RTR-A or RTR-B is the "correct" one to announce the
net? You have a scenario called "segmentation", and there's no easy
resolution.

If there are no "Host1"s you use SW-A as a simple bridge, a little like
using an IMC McBasic media converter, and then you need some "poisoning"
feature, like their "LinkLoss".

If you have both "Host1"s and "Host2"s, which ones would you like to not
have any connection to the outside world?

Regards,
Peter


On Tue, 2008-01-29 at 14:16 +0100, Gabor Ivanszky wrote:
> Hi Whisper,
> 
> thanks for your response.
> 
> probably I didn't explain the issue very clear, I apologize.
> 
> I try to give an other example for the same issue, which is a bit more 
> practical (and more complicated).
> 
> On the attached figure you see a topology which is intended to provide a 
> redundant access for NET Y to the Backbone. Routers A, B and C connected 
> through a common broadcast multiaccess network. Each of them has 1 
> interface in this network and has an IP address from network x.x.x.0/24 
> . They are running OSPF on this network, so they are electing a DR, and 
> everything works fine until now.
> 
> Now the link between the two switches goes down, so the broadcast domain 
> is separated into two parts: one part with router A and C, and one part 
> with only router B.
> A & C elect the new DR. BUT router B can not see any OSPF neighbors in 
> network x.x.x/24 anymore, so it became DR too, and starts to announce 
> network x.x.x/24. If router B's announcement is more attractive than 
> router A's, than probably it will cause a blackhole, and NET Y became 
> unreachable.
> 
> I could be wrong, because I don't know OSPF path calculation algorithm 
> in detail, but I assume it is somehing like this:
> 
> NET Y is attached to routerC -> routerC is attached to  x.x.x/24 -> 
> x.x.x/24 is attached to router A and router B => if router B's metric is 
> better, then we have a blackhole
> 
> probably I am overlooking something really simple...
> 
> many thanks for your time
> cheers,
> Gabor
> 
> 
> 
> 
> 
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/