Harold,

Good info!!

I was going to dig in and try to find out the reason for this behavior and you've saved the effort :-)

Regards,

Sundar

Thank you! Have you thought any other way to make it possible, maybe changing something on the routes that the route reflector receives to send to its clients???

Any idea?, thanks a lot!

Andres,

Unfortunately, the only way I know to avoid this issue is to get rid of the RRs by doing a full ibgp mesh :o(

With VPNv4 it is a bit different since a prefix is made of the RD+ipv4 prefix. The VPNv4 prefix is therefore considered unique by the RR if its combination of RD+ipv4 prefix is unique and therefore the same prefix with different RDs are not compared by the RR as they are considered different.

The add_path draft would have offered a similar solution for IPv4 but I don't think it is likely that we will see it in any production release anytime soon.

Regards,

Hi Harold,

Just to consolidate the whole info, accordingly if there exists 2 VPN routes for the same network and for the same customer (same RD) thus it will have the same effect on the RR and it won't advertise both with no work around (although this is practically a weired case), is my assumption correct.

BR,

Mohammed Mahmoud.

Mohammed,

You are absolutely correct. This is why it is recommended to use different RDs when a customer site is attached to two or more PEs and you want to achieve load-balancing for traffic to that customer site.

Regards,

Hi Harold,

Agreed, thank you very much for the confirmation.

HTH,

Mohammed Mahmoud.

Hi,

There might be a chance in a normal IPv4 iBGP environment. The idea is to have two RRs - normally the case anyhow.

As those two RRs are independantly doing path selection, they might come to a different best path for a prefix. Thus two "best" pathes would be sent to a client.

The tricky thing is to get a stable environment with the desired results, without breaking BGP design rules, like "Don not modify Loc Pref internally!".

One might use communities to set weigth on the RRs to prefer a specific path.

Just an idea - to be discussed.

My feeling is though it might be too problematic from an operational point of view.

Regards, Martin