OSPF as CE-PE Routing Protocol

[nadeemr]

Hi,

A customer has about 40 sites in MPLS VPN. All sites are stub and have point-to-point connections to Service provider PE's. Few of these sites have dual point-to-point links to PE's. The customer wants to use OSPF as CE-PE routing protocol. Can someone guide what should be the optimal OSPF topology for this network and how the OSPF areas should be design?

[pkhatri]

Howdy,

In a setup such as this, where you are using rfc2547bis VPNs, the use of multiple OSPF areas is not going to buy you much.

My advice to you would be to place each link in OSPF area 0.

Unless you are using sham-links, all OSPF-learned routes from one CE will appear in a CE connected to a different PE as type-3 LSAs...

I have used this design for a few of our customers without any issues.

The use of non-zero areas causes issues when there are backdoor links between sites. In your case, that does not appear to be not an issue, though..

Hope that helps - pls rate the post if it does.

Regards,

Paresh.

[nadeemr]

Thanks Paresh. What I understand from your advice is that I should put all point-to-point WAN links b/w CE and PE in area 0. In this way CE's will see routes from other CE's as inter-area routes.

What about LAN interfaces of CEs. Do I need to configure them for OSPF or not? There might be a couple of sites where CE will have 2-3 static routes pointing to a L3 switch connected on LAN.

regards,

[pkhatri]

Hello,

Yes, put all of the p2p WAN links between PE and CE in area 0. The routes from other CEs will appear as inter-area regardless of which area you use. However, using area 0 simplifies things and the use of multi-areas does not provide any advantages in this case.

You don't have to run OSPF over your LAN interfaces. You could simply redistribute all connected and static routes into OSPF.

However, if you wish to dynamically learn the routes connected to your L3 switches, you can extend yur OSPF process out to the L3 switches. But the redistribution option is far simpler...

Pls do remember to rate posts.

Regards,

Paresh

[nadeemr]

Hi Paresh,

In case the customer wishes to run OSPF b/w CE and L3, should we also configure them in area 0 or use a separate area?

regards,

Nadeem.

[pkhatri]

Hello Nadeem,

In that case, I would run these links in a separate area (use different areas at each site in case you end up having backdoor links between them some day).

The advantage of doing so is that you can apply route summarization for these areas.

Hope that helps - pls rate posts that help.

Regards,

Paresh

[nadeemr]

Hi Paresh,

If I put all CE's in area 0, don't you think the size of area will become too big. It can also have scalability issue if customer wants to add more sites in future. Secondly, it will also result in too much load on CEs. Whenever there is a topology change all CEs will have to recalculate OSPF table. Moreover, LSA flooding will also be a problem in such a big area 0.

Please advise.

regards,

[pkhatri]

Hi Nadeem,

That will not be the case. When OSPF is used as a PE-CE protocol in a rfc2547bis network, things work a lot differently.

Consider the case of a single PE-CE link running as area 0 (ignore the link to your switches for the moment). Also assume that only one CE is attached to this PE:

- there will only be two type-1 LSAs within this area, one for the PE and one for the CE. This is regardless of how many other PE-CE links running OSPF are present on other PEs. This is because the rfc2547bis network is considered as an OSPF super-backbone when running this way...

- routes from other PEs will be learnt as type-3 LSAs.

- every time there is topology change somewhere, only the type-3 concerning that will be flooded to all other PEs. The change of a type-3 LSA does not cause an SPF calc

If you extend this case to where you have x CEs connected to a PE, each CE/PE will have x+1 type-1 LSAs, one for each CE and one for the PE.

So you see, this is a different thing to a 'normal' OSPF setup...

So I stand by my previous recommendations...

Pls do remember to rate posts.

Paresh.

[nadeemr]

Hi Paresh,

The customer is starting with 40 sites. The network might grow to 200+ sites in few years. Do you still think we can put 200+ routers in one area?

regards,

Nadeem.

[mheusinger]

Hello,

as already stated in the previous posts your routers will not have a topology database reflecting 200 routers in one area. Each location will have only the routers in that location plus the PE(s) in the database. So there should be no issue arising from adding more sites.

The main concern with many routers in one area comes usually from CPU and memory usage due to many router LSA and stability issues.

The stability issues are not of such a concern here with respect to CPU load, because most networks will be learned through LSA3 which will not trigger a SPF calculation. This means even if a site of the 200 will go down the majority of your routers will not have to do SPF, so the CPU impact would be minimal. In case you are really worried you could also apply a "nice" IP addressing scheme and summarize all networks in a site into one route. Doing this would further minimize the impact on other sites and routers there. From a site router perspective the other sites are in a different area - this is what the MPLS VPN PE will announce to them through LSA type 3, which means to them INTER area routes will be learned.

So in short:

A) you are not placing 200+ routers in one area (though technically they are configured with area 0) with respect to your OSPF database.

B) The network you are about to setup will scale to this size and no problems should arise from increasing the number of sites - assuming proper MPLS VPN configuration.

Hope this helps! Please rate all posts.

Regards, Martin

[pkhatri]

Hi Nadeem,

Just to make you feel better about this :-), I have deployed a 400-site network using this and it is working quite successfully.

So go ahead and do the same and you will find that you will not have any issues...

Pls remember to rate posts.

Regards,

Paresh,

[dknov]

Hi guys,

It's an MPLS forum, so the following question is a bit off topic, but since this discussion had already started I will carry on :-)

Why do you say that Type 3 LSAs do not trigger SPF recalculation? I understand that all the routers in a given area synchronize their link state databases through exchanging LSAs and they all hold the same view of network topology. However summary LSAs are also part of this view (LSD that is), so why do you say that when new summary LSA is received, SPF recalculation is not triggered?

Thanks,

David

[pkhatri]

Hi David,

Yes, it is treu that type-3 LSAs are part of the LSDB. However, there are a couple of distinctions to be aware of here:

- OSPF uses only type-1 and type-2 LSAs for topology determination. These LSAs provide the true link-state natiure of the protocol - they provide information on the actual links between routes.

- types 3,4 and 5 are used in more of a distance vector fashion - the only provide information on distances (metrics) to routers and networks

Therefore, when type-1 and type-2 LSAs change, an SPF computation has to take place. Whereas, when any of the other three LSAs change, all that needs to be done is to update the routing table. Section 13.2 of RFC2328 describes this process quite well...

Hope that helps - pls rate the post if it does.

Regards,

Paresh.

[mheusinger]

Hello David,

maybe another approach to understand this behaviour: The whole idea of creating multiple areas is to reduce CPU and memory requirements for an OSPF router.

So an area internal router will only learn the topology of his area (LSA1 and LSA2). The ABR announces all the rest of the networks as "being reachable through me" (LSA3). As one wants to have the distinction of a network being connected to the ABR in an area, and therefore being part of the topology, and the networks learned through another area an new LSA type was used - LSA3. This way an ABR can announce: There is a new network reachable through me - BUT the topology has NOT changed in this area. As the topology has not changed no area internal router recalculates it. They just will accept the fact that there is yet another network reachable through the ABR and insert it into the routing table.

So CPU consumption is reduced - no SPF calc - yet still connectivity preserved.

Hope this helps! Please rate all posts.

Regards, Martin

[juttaro]

My advice would be to run BGP & Static where applicatble..