[c-nsp] Peering between route reflectors

[Cydon Satyr]

@JC Cockburn

Why not is always an interesting question!

I don't think there's any difference in terms of functionality. It's just
that I've seen most engineers automatically put down peering between two
RR, without explanation or clue why. Given how powerful single RR boxes are
today there isn't any good reason to say one way is better than the other,
IMHO. You save some memory from not having routes reflected from other RR,
which may or may not be important.


In fact I was looking at how would a design look like where you have 4
POPs, 3 would have 2 RR, last POP only one RR. None of RR are in forwarding
path. I created four clusters  - each with two RR, except last POP cluster
with only one RR (at the moment). There would be full mesh between POPs,
but no peering between two RR is same cluster, which made sense to me.

Hence, I got to ask the question I originally asked - why should two RR
mutually peer at all, given certain conditions - all clients peer with both
of them and RR are not in forwarding path.

Regards




On Tue, Apr 8, 2014 at 5:17 PM, Vitkovský Adam <adam.vitkovsky at swan.sk>wrote:

> > Gert's got it, a network with a massive # of clients may justify
> hierarchical RRs
> > although the design might be a throwback to the days of small boxes with
> > minimal memory. Today's control plane only systems can scale to hundreds
> if
> > not thousands of clients. Hierarchy might be useful if the second tier
> RRs are
> > in the forwarding path and need to conserve resources for something other
> > than BGP.
> >
> > Just curious, anyone out there using hierarchical RRs?
> >
> > Lee
> >
>
> Yes you are right the complex combinations of topology-based and
> multi-planar address-based RR designs are the state of the art legacy from
> days where engineers had to fight back the lack of memory in routers while
> distributing millions of VPN prefixes around the globe.
> Though I think the topology-based RR models are still applicable even with
> modern gear.
> Would these be hierarchical RR designs? No, not really.
> So unless the partial-mesh between nodes in clusters becomes unmanageable
> in a particular region you don't need a RRs hierarchy.
>
> address-based RR = set of RRs distribute only a subset of all routes.
> topology-based RR = set of RRs serving a particular geographic area.
>
>
> adam
>
>
>
>
>
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>