PE-CE OSPF E2 routers preferred over IA!

[m_almoez1]

Hi all,

I built topology like this :

    __________(MPLS)___________

    |                                                     |

R1-------------------Backdoor------------R2

R1 and R2 run OSPF area 0 on all interfaces except the one connected between of them. i configured the backdoor link in area 100.

What i know OSPF should select the internal routers over external, but this is what happening with me, E2 is preferred over IA

when i configured the link (Backdoor) between R1 and R2 to be in area 0 too, it gives me the needed result. 
I got confused when i follow the flow with the DN so i didn't get it until now.

Anyone can help me regarding to this?

Thanks in advance!

[Peter Paluch · Accepted solution]

Hello,

You say yourself that R1 and R2 have all their interfaces in Area 0 except the backdoor link that lies itself in Area 100. Now, when the MPLS connectivity is down, notice what happens: You have two partitions of Area 0 on each side, interconnected by a link in Area 100. To OSPF, this is an invalid configuration. Remember that OSPF requires that Area 0 is contiguous at all times and cannot be partitioned. If two partitions of Area 0 are interconnected by a link that lies in a different area, OSPF will ignore this interconnection. More precisely, both R1 and R2 will take their respective Area 0 internal routes and inject them into Area 100 as LSA-3. However, because R1 and R2 are ABRs, they can use only LSA-3 received over Area 0 when computing their routing tables. This backdoor link will therefore be effectively disabled - as if it was not even there. You will have an OSPF adjacency over that link, but routing over that link will not take place.

I do not even believe that this thread's heading is correct: E2 were not preferred to IA routes. Rather, E2 routes were the only ones that were available. These routers never considered the backdoor link for IA routes because they ignored each other's LSA-3.

Backdoor links generally assume that both sites interconnected by the backdoor link are located in the same area including the backdoor link itself. If both sites use a different OSPF area then the backdoor link must be in Area 0 itself - otherwise it will be ignored just like you have experienced yourself. This would pose additional requirements on the area design of the individual sites: You cannot just place a backdoor link in Area 0 without considering existing Area 0 locations and making sure they all together form a contiguous Area 0. If the backdoor link is to be located in Area 0 then each site is required to have its own Area 0 and the backdoor link must basically be connected between ABRs on both sites.

Best regards,
Peter

[m_almoez1]

Thanks Mr. Peter. The way you elaborate is very simple and nice.

Suppose we have loopback interfaces on both R1 and R2 which are on area 20, which path they should follow with same configuration as above.

Because it goes through the backdoor instead of going through the super backbone area.

[Peter Paluch]

Hello,

Assuming that the backdoor link is in Area 100 and the loopbacks are in Area 20, the traffic between loopbacks should go over the superbackbone. You are writing that the traffic goes through the backdoor link instead, which is - according to the information available so far - surprising.

Can you perhaps post the entire configuration of R1 and R2?

Best regards,
Peter

[m_almoez1]

Thanks Mr. Peter, sorry for replying late.
 

R1
router ospf 100
network 1.1.1.1 0.0.0.0 area 20         ========> Loopback interface
network 10.1.2.1 0.0.0.0 area 0         ========> Interface connected to PE-1
network 10.1.20.1 0.0.0.0 area 120   ========> Interface connected to R2 (Backdoor link)

R2
router ospf 100
network 2.2.2.2 0.0.0.0 area 20          ========> Loopback interface
network 10.1.5.1 0.0.0.0 area 0          ========> Interface connected to PE-2
network 10.1.20.20 0.0.0.0 area 120  ========> Interface connected to R1 (Backdoor link)

[Peter Paluch]

Hi,

Thanks for the illustrative depiction of your topology!

I have replicated this setup but I am unable to confirm your observations. If all links and routing protocol adjacencies are up then all routing goes through the MPLS superbackbone, including the routing between the loopbacks in Area 20.

If exactly one PE/CE link is brought down, situation starts to be more complex. The CE router that has its PE/CE link up (let's call it CEup) still considers itself to be an ABR, and is therefore allowed to advertise networks from one area to another. Therefore, it will advertise all its known OSPF networks via the backdoor link to the other CE router. The second CE router having the PE/CE link down (let's call it CEdown) no longer considers itself an ABR, as it does not have any active connection to Area 0, and therefore starts to behave as a simple internal router, processing any LSA-3 it has received. CEdown will therefore know all routes advertised from CEup. However, CEup will stop seeing any routes from CEdown, because CEdown is no longer an ABR and is not allowed to advertise routes from one area to another - it must not advertise its networks in Area 0 and Area 20 into Area 120 where the backdoor link is.

If both PE/CE links are down then none of CE router will see the other CE routers' networks because the CE routers are no longer ABRs and must not advertise networks from other areas to each other.

This entire explanation assumes that the backdoor link is placed into a non-backbone area. If the backdoor link was placed into Area 0 then the situation would be radically different - so different that all routing would take place over the backdoor link, and the MPLS superbackbone would be used only if the backdoor link was down. This could be solved by configuring an OSPF sham-link in the MPLS superbackbone - but at this point, this is fetching too far. Let's get your scenario working right first.

So once again, are you saying that with all links up, all routing protocol adjacencies up, all redistribution configured properly, traffic between loopbacks on your R1/R2 goes through the backdoor?

Best regards,
Peter