MPLSoDVMVPN Scalability for L3 MPLS VPN and VPLS

[canero]

Hello,

The choice of IGP for Dmvpn effects the scalability of Dmvpn design. In order to extend an MPLS network that provides L3 Mpls Vpns through a service provider, L2 VPLS and L3 VPN (i.e mpls over vpls or mpls over mpls) through the SP seems to be the suitable choices. For security the use of DMVPN tunnels for each solution we think MPLSoDMVPN may be a good solution for both vpls and mpls.  Only disadvantage seems spoke-to-spoke traffic should go through dmpvn hub, which may seem an advantage as it provides a more predictable path for spoke-to-spoke traffic.

1. If the service is provided through a L2 VPLS service, it seems that all the WAN PE Routers will seem to be connected to a big LAN switch, so if RIP is used for example a size of 1000 sites, we think it will create a huge routing update. Also OSPF for LAN, or EIGRP may create convergence time or cpu load for a Lan of 1000 sites or more.
2. If the service is provided through an L3 Mpls VPN Service, the PE routers will be seperated through a multihop routed VRF. May this choice has some points of concern? MTU size seems one point. For the scalability of this solution BGP table size, use of route reflectors is recommended.

If scalability and performance is the main design issue, which solution is more scalable and brings less burden on the PE devices?  May there be other choices or points that may be important?

https://learningnetwork.cisco.com/thread/2789 --> This thread  gives the limitation of the regarding the number of Mac addresses and bgp prefix number limitation, but routing may the IGP choice for LDP also important if  LDP is to be transported through mGRE tunnels?

Thanks in Advance,

[maayre]

Not really an answer to the question you actually asked but have you considered Carrier Support Carrier (CSC)? It allows you to leverage a providers MPLS backbone as your own (inside a provider MPLS VPN / VRF) avoiding the need to create an overlay network.

If you are planning on having all traffic go in and out of a hub site you will risk;

1) geographical redundancy (ie if your hub site has issues all spokes are down)

2) traffic duplication, you will need more resources to send the same amount of traffic

Personally I don't agree that the path is any more predictable in either scenario, it relies on provider routing and is pretty transparent to your edge routers.

On the other hand you do want to find a way to avoid configuring "n(n-1)/2" tunnels.

HTH,

Matt

[maayre · Accepted solution]

PS - DMVPN is supposed to avoid traffic being sent through hub for spoke to spoke communication, it looks like you said it will force it through the hub

[canero]

Hello Matthew,

Thanks for the reply, seems a good solution,  the CSC seems a good way of carrying an mpls backbone over another SP MPLS backbone, which provides spoke to spoke availability.

In case of scalability, can we think it brings less overload than Dmvpn, and when we look at the supported platforms ASR9000, 7200 and 7500 routers are supported for this feature.

Best Regards,