How are Route reflectors used in an MPLS L3 VPN network

[carl_townshend]

Hi Guys

In a typical ISP network with lots of PE's and offering services such as L3 VPN with multiple customers.

I see most the time the PE's are all neighbours with each other.

However I read this is not scalable and so RR's are used, in which case would the PE's not run ibgp to each other? would they run ibgp to the RR? 

If sharing VPNv4 routes, would the RR's have to advertise the vpnv4 routes and pass all this info on? would the route reflectors need to run multiple address families and route targets etc like the PE's would typically do?

What would a typical design look like? and at what point would you need a route reflector normally?

Also, do most providers have the RR's in the transit path or have them separate? if separate, you would still need a physical link between each router?

[MHM Cisco World]

-RR will reflect the VPNv4 only so no need other address family 

-if you have ten PE then you need 9 iBGP between one PE and other PE's
instead you can use RR which in this case you need one iBGP to RR and RR will reflect the VPNv4 prefix to all other 9 PE's

-the RR no need to be in DATA path between two PE 
- the RR must have link to each PE but this link used to exchange the VPNv4 establish iBGP but not for  pass DATA traffic.

- the design with one RR look like hub and spokes 

[carl_townshend]

Hi

if route reflector is not in the data path, then if you had hub and spoke the with RR as the hub,,how would you route from spoke to spoke if you only had one physical link to the RR from each spoke?

[MHM Cisco World]

 

I Know you will ask this Q'

The hub spoke for ibgp and vpnv4

And for data sure there must data path between two PE want to connect with each other.

[carl_townshend]

thanks for the response, so would all the PE’s have some other physical data path then ?

what are the benefits here? Is it just scalability so you don’t have to have lots of neighbours?

Are most of these ibgp neighbours directly connected generally ?

[MHM Cisco World]

thanks for the response, so would all the PE’s have some other physical data path then ? Yes

what are the benefits here? Is it just scalability so you don’t have to have lots of neighbours? we talking inside SP so we talk about tens or more neighbors so for admin this nightmare

Are most of these ibgp neighbours directly connected generally ? if there is RR no need for full-mesh iBGP 

[Harold Ritter]

Hi @carl_townshend ,

> would the route reflectors need to run multiple address families and route targets etc like

> the PE's would typically do?

The route reflector could have multiple address family (ipv4, ipv6, vpnv4, vpnv6, etc), but large service providers will generally dedicate route reflectors for each address family for scaling reasons. Route targets don't come into play on the RR. The RR will keep all routes regardless of the route target they are tagged with and just reflect them to all route reflector clients. The RRC are responsible to drop the routes not matching local route target imports.

Regards,