thanks for your reply Giuseppe.

"enabling MP BGP on the CE can have a huge impact without filtering." Can you emphasize on this a bit?

My understanding is that MBGP will not install the route in its BGP database unless it has a matching route-target in its configuration, which will make it so only the configured VRFs on the CE router will be installed.

"if FR multilink is not supported in the CE platform then you can also think to use the links as separate L3 links each with one FR subif for each VRF." If I understand that correctly, this is what we currently have, how is this helpful wanting a bonded T1?

Hello Sergio,

1) enabling MP BGP on the CE can have a huge impact without filtering.

vpnv4 prefixes will be imported if RT matches but before all of them are received so how much memory there is on the CE ?

2) I was suggesting to use a L3 load-balancing that can lead to the use of all T1 links up to 6 or 8 (ios dependent).

they will not be bonded but they will be used all. Load-balancing will be flow based and less smooth then with PPP multilink or FR multilink. It depends on your requirements: if it is : we want to use all the available links also a L3 load-balancing is acceptable.

Hope to help

Giuseppe

1. are you sure all of them are received?

let me simplify.

If I have 2 PE routers and they only share 1 vrf, they will discard routes that do not match that VRF. There is an implied automatic route-filter based on the route target:

http://www.cisco.com/en/US/docs/ios/12_3t/mpls/command/reference/mp_a1gt.html#wp1015775

In regards to your 2 suggestion, we are aware of per flow load balancing and this is fine is some scenarios, but we are looking to aggregate our b/w and do per packet with the help of MLFR or MLPPP.

Hello Sergio,

I was meaning this:

Use the bgp default route-target filter command to control the distribution of Virtual Private Network (VPN) routing information through the list of VPN route-target communities.

When you use the no form of this command,

>>all received VPN-IPv4 routes are accepted by the configured router. Accepting VPN-IPv4 routes is the desired behavior for a router configured as an autonomous system border edge router or as a customer edge (CE) BGP border edge router.

I was pointing out that prefixes are received before and then imported if RT matches.

However, the total of vpnv4 prefixes accepted by your PE will be passed to the previous CE now promoted PE in option 1 scenario.

Hope to help

Giuseppe

They will not be stored into memory if the RT does not match, so thats not relevant in this case.

Any other reason not to extend MBGP/MPLS to the CE router?

Hello Sergio,

I don't see any other drawback on extending MBGP/MPLS to CE

Best Regards

Giuseppe