[j-nsp] Virtual route reflector physical layout

[Mark Tinka]

On 6/May/16 17:53, Clinton Work wrote:

> NANOG65 had a couple of useful presentation (video presentation links
> via the agenda) about large scale RR deployment and shadow vRRs.  The
> Juniper shadow vRR technique is interesting as you can deliver all your
> vRRs from two or three central data center locations.  
>
> https://www.nanog.org/meetings/nanog65/agenda
>
> https://www.nanog.org/sites/default/files/tuesday_tutorial_bothra_large-scale-bgp.pdf
> https://www.nanog.org/sites/default/files/wednesday_general_szarecki_fire-water.pdf

Ouch, that seems overly complex, to be honest. And fragile.

I think BGP ORR would be a much more elegant solution than a "shadow" RR:

   
https://tools.ietf.org/html/draft-ietf-idr-bgp-optimal-route-reflection-11

AFAIK, Cisco have an implementation in IOS XR for the ASR9000. I'm
pushing them hard for IOS XE/CSR1000v as well. I think Juniper were
planning a release around Junos 16.

I don't generally like to centralize things, and certainly not RR's.
Complete backbone link failures where ever your centralized RR's are can
take out your entire routing domain very quickly.

Virtual RR's makes the deployment of RR's cheap, because servers are
cheaper than routers, pound for pound. So why be concerned about the
(minimal) costs of decentralization if it guarantees snoring at 3AM?

Mark.