BGP redundancy over frame relay

mfieldhouse

I have a lab in dynamips - 3 routers, R1, R2 and R3 connected to a frame switch. Point to point PVCs have been created between R1 + R2 and R1 + R3. R2 and R3 are connected via a switch.

R1 is in its own AS, R2 and R3 in the same AS, so iBGP between R2 + R3 and eBGP between R2 + R1 and R3 + R1

The 172.17.32.0/22 network sits behind R1, 172.16.32.0/22 behind R2+R3.

Say traffic is routing over R1 to R2 to reach 172.16.32.0/22, the problem is that when shutting down the serial interface on R1 to simulate a WAN link failure, the BGP session terminates fine on R1 - as it loses it's connected route to the network between R1-R2. But because R2 is connected to the dynamips frame relay switch, it's interface stays up/up - and the BGP session does not terminate until the hold time expres, resulting in a relatively long convergence time. Eventually traffic does reroute correctly over R1-R3.

Now, my question is - is the problem with R2 not recognising the link to R1 is down because of the dynamips way of doing things, because the way I have the lab configured or because this is how it works in real life.

Eg: in a real network would the serial interface go up/down on R2 when R1 serial is shut.

Let me know if you need clarification on anything.

Thanks!

kryolla

Am I understanding this correctly R1 is connected to R2 and R3 via a FR point to point link and R2 and R3 is connected via ethernet?

if the sub-interface is a point to point frame relay between R1 and R2 by shutting down the sub-interface on R1 it should bring down the PVC status on R2 and line protocol status for the point to point link. I think BGP fast external fallover works via the line protocol status so it should of brought down the BGP session and not wait until the dead timer expires. If it is multipoint then it works differently.

Eg: in a real network would the serial interface go up/down on R2 when R1 serial is shut.

line protocol status on the physical interface for frame relay is dependent on its connectivity to FR switch and not the other end. As long as it keeps getting LMI from the FR switch it will be in an up/up state. Same thing with ethernet, the port will stay up as long as it has connectivity to the switch. There is no end to end keepalives just to the directly connected switch or FR switch. When you start running routing protocols, they have their own end to end keepalives called hellos

mfieldhouse

Am I understanding this correctly R1 is connected to R2 and R3 via a FR point to point link and R2 and R3 is connected via ethernet?

Yes, that's correct and your explanation of LMI makes sense.

I've just started the lab back up, traffic from R1 to the R2/R3 network is routing via R3. I shut down the R3 point-to-point subinterface. The corresponding subinterface on R1 stays up/up and the PVC active. The PVC on R3, even though its subinterface is shutdown is also showing active.

As soon as I shut the subinterface down on R3, it goes "%BGP-5-ADJCHANGE: neighbor 172.21.0.5 Down Interface flap" but because as far as R1 can see, everything is ok, BGP does not go down.

Any ideas or could this just be a Dynamips "bug"?

kryolla

The PVC on R3, even though its subinterface is shutdown is also showing active

dynamips bug but I dont work with emulators