Large VPLS WAN - Customer routing design

[craig.cartlidge1]

Hi, 

I work for a large organization with around 400 edge sites. these sites are connected via leased lines in a hub and spoke fashion back to  out primary data center.

We run OSPF as the IGP to route between sites, the Datacentre is OSPF area 0 as this is where all the hub sites leased lines terminate. Each hub is an area. A pretty simple/standard ospf setup.

Management are wanting to move to a VPLS WAN solution that will save alot of capital as opposed to leased lines.

my understanding of VPLS is that it appears to the customer as a single LAN segment.

My question is, what kind of routing setup from a customer perspective are people running over VPLS and why?

Im guessing it wouldnt be best practice to have all Routers L3 adjacent with all other routers in the WAN.

Any case studies or real world examples of an enterprise VPLS WAN from the customers perspective would be very much appreciated. 

[Jose Jara]

Hi Craig,

for this number of sites I recommend you to choose MPLS/VPN instead of VPLS, as it scales much better. 

Best Regards,

Jose.

[Hitesh Vinzoda]

hi Craig,

VPLS solution is sometime provided point to point or point to multipoint fashion, other variation will be 

SP will provide port to port circuit which what come in from B end will be sent to a port on their NTU at A end ie. DC.

They may also do port to VLAN which is take any ingress traffic on a port at B end and tag it as VLAN when presented at A end i.e. DC

You may take point to point links with tagged VLANS from the SP i.e. one VLAN for each site and use it for creating adjacency between A & B end.

VPLS essentially doesnt means that you will have LAN to LAN traffic. VPLS will be preferred solution for a SP as they dont have to manage your routing and table and it will be easier for you not to contact SP in case you see any issues with routing within the MPLS cloud.

You may ask for details to your SP for above details and is a much neater design.

HTH

Hitesh

[Peter Paluch]

Hi Craig,

I tend to agree with Vitesh - from the perspective of the SP, VPLS is much more comfortable than MPLS L3VPN. SP will not need to deal with the PE/CE routing protocol on each site, with potential redistribution to/from BGP, carrying your routes in BGP and having them properly sorted to VRFs at his PEs, optionally providing you with OSPF sham links if you require the use of backdoor links, does not need to coordinate the IP addressing of PE/CE links on individual sites, etc. All these facts make VPLS attractive to a SP. Of course, this is not really your problem because you are a customer and you basically do not need to care.

VPLS as a service can be very interesting to you as a customer, too. First of all, instead of 400+ subnets just for the PE/CE links, you now have only a single backbone-like network. Even with private addressing, 400+ subnets is going to consume a lot of management resources, and is going to clutter your routing tables unless you do some clever things like OSPF Prefix Suppression. Moreover, you do not coordinate the addressing and the PE/CE routing protocol with your SP anymore - you simply deploy the addressing and the routing protocol anyway you like. All the issues with OSPF/BGP redistribution are simply gone, and so is the need for advanced features like the OSPF sham links. Should you decide to deploy IPv6 in future, you do it in a snap of a finger, without ever asking your SP for a particular support.

Having 400+ edge routers on a common subnet obviously poses some challenge to the way the routing protocol will be deployed. Assuming the VPLS will be provided with full visibility between all 400+ sites, running OSPF in a broadcast network mode would be ridiculous. In that case, it is possible to force OSPF to operate this network as a NBMA-type network, making sure that there would only be OSPF adjacencies from spokes to the hub, and not between spokes themselves. OSPF could even be forced to construct the routing tables so that not only the OSPF adjacencies would be of the hub-and-spoke type, but also the data traffic would be routed in a hub-and-spoke fashion without direct spoke-to-spoke flows. Either option is available (OSPF NBMA network type would provide hub-and-spoke OSPF adjacencies while allowing for direct spoke-to-spoke data flows; OSPF Point-to-Multipoint NonBroadcast network type would provide for hub-and-spoke adjacencies and hub-and-spoke data flows).

Before deciding on how to deploy your OSPF over this network, though, you should find out whether the provider is going to provide the VPLS service with full visibility between individual sites, or whether this VPLS is going to be a hub-and-spoke service.

My two cents...

Best regards,
Peter