How do most ISPs build there MPLS VPNs, do they fully mesh all PE's ?

[carl_townshend]

Hi All

I am curious how most ISPs build there MPLS VPN networks.

Do they fully mesh all there PE routers together via the core and use MPLS tunnelling to ensure that all routes can reach each PE due to the label only seeing the next hop PE IP?

What do most providers use as the IGP to share the BGP reachability info so BGP routers can speak to each other?

cheers

[Giuseppe Larosa]

Hello @carl_townshend ,

a fully mesh is not possible for the fact the networks are so big with so many PE nodes in terms of hundreds or even thousands of PE devices.

The PE nodes are usually deployed in pair each of them connecting to the core to a different "forwarding plane" .

Each of these two forwarding planes provide two indipendent end to end paths.

the two PE nodes form a POP point of presence and they are deployed in each town where the ISP is present.

Each POP is connected to a core POP where there are at lest two P routers one for each forwarding plane.

The most used IGPs are IS-IS and OSPF because they are link state and are the only two protocols that support MPLS traffic engineering.

What is important is to advertise each PE loopback address so that MP BGP sessions with BGP route reflectors can be built.

This solves the full mesh issue also on the control plane as BGP RR can reflect /propagate routes received from a client to other clients and to non clients.

A full mesh of  i MP BGP sessions is required between the RR nodes to make possible for each PE node to receive interesting VPNv4 routes coming from any PE node even in a different region.

Hope to help

Giuseppe

[carl_townshend]

Hi Giuseppe

Thanks for the reply

So when you say the 2 PE's would be on different forwarding planes what do you mean?

Do they mostly use ASR routers in the POP?

What would be acting at the Route reflector, the core?

Do you have any diagrams that I could look at to make it clear?

Many thanks for your help

Carl

[Giuseppe Larosa]

Hello Carl,

the PE routers are usually ASR9000 routers for powerful PE nodes ASR1000 or some models of Metro ethernet switches like ME3600 ME3800 can be used for less powerful PE nodes.

The route reflector servers are usually implemented in dedicated nodes and they typically are ASR 1000 with added RAM 16 GB or more.

The two PE nodes of one POP connect to two different CORE nodes this is how the connection to the two indipendent forwarding planes is performed.

The topology can be seen as two different planes /floors so that when a user stream enter in a PEn1 it will travel on floor 1 until the remote POP and will exit from PEm1 using floor/plane1 resources.

If anything happens on the floor1 end to end path , the user traffic will be processed by Pen2 and will reach PEm2 travelling on floor/plane 2 links/nodes.

This topology creates two indipendent "forwarding planes" if anything happen on one plane/floor the other plane is not affected.

This happens also for temporary traffic engineering re-routes of tunnels by using some attributes associated to each link a sort of color or affinity group that tells which LSPs made by LDP and which tunnels made by RSVP-TE can use a specific link.

The idea is that once the user stream is placed in one plane it cannot be rerouted on the other plane/floor and it is carried to the outgoing POP if possible.

Hope to help

Giuseppe

[carl_townshend]

Would you have any diagrams of this setup at all ?