OSPF over VPLS

[juanlarriega1]

Hello Engineers;

I have a question on setting up the type of OSPF network for Ethernet interfaces facing the VPLS/WAN network. I've read recommendation to change the OSPF network type from broacast to Point-to-Multipoint and Point-to-Multipoint Broadcast, I like to know what will be the best configuration for ethernet where sites have different bandwidth to the VPLS network. Some sites connect to the VPLS network using FastEthernet and some other Gig and 10 Gig connections. Is OSPF cost should be consider on  a VPLS network? Does proxy-arp need to be disable on the ethernet interface?

Thanks;

Juan

[Masoud Pourshabanian]

Hello,

I do not see any benefits over using Point-to-Multipoint. VPLS simulates the function of a switch, so broadcast network type works fine. Point to multipoint is usually used in Frame relay networks(partially designed networks). If your service provider allows multicast traffic, broadcast type is fine.

you should be concern about OSPF metric If you have redundant paths toward your other branches; otherwise, the default metric is fine.

I usually disable proxy ARP. If subletting is correct in the network, there is no need to enable it.

Hope it helps,

Masoud

[Rolf Fischer]

Hello Juan,

I generally agree on what Masoud said.

However, depending on your requirements, the Point-to-Multipoint [Non-Broadcast] network type can be an interesting alternative to solve certain design problems. If you have several neighbors on an OSPF Broadcast interface, the cost of this interface is inevitably the cost to each of those neighbors - you cannot assign different costs to different neighbors. Point-to-Multipoint, on the other hand, allowes you to configure costs on a per-neighbor basis.

Another advantage of the P-2-MP network type is that it is generally more stable because there are no DR/BDRs which require underlying full-mesh connectivity, and you can create partial meshes which best meet your requirements without such restrictions.

The disadvantages are more configuration effort and, depending on the number of peer routers on the segment, less efficient routing updates and synchronization procedures (that's the reason why network types with DRs were developed in the first place). And some people don't like the hostroutes generated for each of the interfaces; the fact that the actual network(mask) is not advertised can lead to connectivity problems to non-OSPF IP-hosts (for instance BVIs of transparent firewalls or HSRP/VRRP vIPs) within that subnet.

So it really depends on your design goals. If you need to assing different costs for particular neighbors on a segment, P-2-MP may be the solution you're looking for; the non-broadcast option additionally allows you to control which routers become neighbors.

HTH
Rolf

[Masoud Pourshabanian · Accepted solution]

Just adding one point. Rolf, correct me if I am wrong.

Point-to-multipoint is used in hub and spoke topology(partial mesh). OSPF treats Point-to-Multipoint networks as a collective of point-to-point links so you can configure cost per neighbor; however, traffic and routing update of spoke to spoke must go through hub, which is not efficient. If you have mesh connectivity( each location can see others directly), you should use OSPF broadcast.

If your topology is hub and spoke( by nature or DMVPN is used, ....), Point to multipoint can be a good alternative.

Masoud

[Rolf Fischer · Accepted solution]

Masoud,

point-to-multipoint is often used in hub-and-spoke topologies but you can build partial or full mesh topologies as well.

A practical example: We have a remote site connected to the main site via VPLS. Two LAN coreswitches at the main site and a pair of layer-3 distribution switches at the remote site, all in the same OSPF area. One requirement was an active-standby OSPF setup between the sites, and we also wanted to avoid an additional neighborship between the two coreswitches and between the two distribution switches via the VPLS. Sure, you can achieve similar results by changing OSPF costs on several interfaces, but we decided to use the more flexible P-2-MP NB network type. Now every router has neighborships only to the two routers at the remote site and one conection is prefered over the other.

BR
Rolf

[Masoud Pourshabanian]

Thanks for your response.

point-to-multipoint is often used in hub-and-spoke topologies but you can build partial or full mesh topologies as well.

I had considered it in my answer.

If your topology is hub and spoke( by nature or DMVPN is used, ....), Point to multipoint can be a good alternative.

regard,

Masoud

[Rolf Fischer]

Agreed. My example was to show a scenario with point-to-multipoint in a not-hub-and-spoke topology.

[juanlarriega1]

Hi Rolf;

Thanks for your response, On a scenario with two VPLS cloud, one primary and 2nd standby, do you recommend to use P-2-MP NB with 11 remote routers on each cloud?

If you let the VPLS interface default to "broadcast" will you use BFD in conjunction to OSPF?

Thanks;

Juan

[Rolf Fischer]

Hi Juan,

could you elaborate a bit on the topology? How many routers do you want to connect via the two VPLS clouds, how many sites do you have, do you want to avoid equal cost load sharing on some links, etc.

Perhaps you have a simple diagram you could share?

I'm not aware of any restrictions in the interoperabiltiy between BFD and the various OSPF network types, they should all work together well with BFD as far as I know. Maybe somebody else will share his own experiences.

Best regards,

Rolf

[juanlarriega1]

Hi Rolf;

Thanks for taking time to respond to my question. I am attaching a visio diagram. We have 12 remote sites connecting to two VPLS cloud, each remote site has two routers connecting to each of the VPLS cloud, we want to have one VPLS cloud to be the primary and carry all the traffic and have the 2nd VPLS cloud as a backup. We are controlling the OSPF traffic by using OSPF cost of 1 and 100 for the primary and backup links. Some of the interfaces facing the clouds from the remote sites Fa, Gig and TenGig provisioned by the VPLS provider using different bandwidth.

The interfaces facing the VPLS cloud and the link between the two remote routers are on area 0. The connection between the ABR router is a L3 connection.

I am not sure if BFD can help us in a VPLS implementation because VPLS is not a true Layer2 topology. Perhaps you may have some recommendation.

Thanks;

Juan

[Rolf Fischer]

Juan

Thanks for the additional information and the diagram.

You use different subnets for the primary and backup VPLS, right?

From a design perspective, I don't see an advantage using the point-to-multipoint network type here. With network type broadcast all routers on the segement will have the same distance defined by the cost of the local interface, but there are no neighbors advertising the same prefixes on the primary VPLS, correct?

A question we haven't discussed so far is where to place the DR and BDR in such a scenario. This could be an issue during failures in the VPLS because isolated partitions of your network may need to elect new DR/BDR which may affect (re-)convergence in those parts of the networks. In this regard, point-to-[multi]point is more robust, but this is not a very strong argument I guess.

Regarding BFD, let me do some reading first. I'd also really like to hear the experiences and recommendations of other forum members. What are your failure detection requirements, I assume standard OSPF timers (10/40) are too slow?

Regards,

Rolf

[juanlarriega1]

Hi Rolf;

Thanks for your input on the VPLS question. We are using unique subnets for the Primary and Standby VPLS connections. 

The reason for my BFD question is for some reason there is a problem in the VPLS cloud, it would take 120 seconds (2 minutes) before OSPF reconverge again. I am assuming the OSPF node doesn't see the keepalives from its neighbors.

Juan

[Rolf Fischer]

Juan

The reason for my BFD question is for some reason there is a problem in the VPLS cloud, it would take 120 seconds (2 minutes) before OSPF reconverge again.

You can (and you should) change the OSPF default timers (30/120) without having to use BFD. BFD allows much faster failure detection and is less CPU-intensive but I don't think you need such aggressive timer values. Have a look at the configuration in this discussion, he has defined 1s Hello timers and 5s Dead Interval. Would that meet your requirements too?

Rolf

[P.S.: Keep in mind that timers have to match among all neighbors.]

[juanlarriega1]

Thank you Masoud and Rolf for your valuable information.

I am not familiar with VPLS but I've read that VPLS is built as a full mesh topology, that was the reason for asking what type of OSPF network should be configured on the interfaces facing the VPLS cloud.