🚀 go-pugleaf

On Sun, 2 Apr 2000, Dave Cooper wrote:


> Depends on many factors, including physical and logical topology.  Also,
> clients dont have to be downstream from the RR. You may have an acting
> RR that provides routes to a client but never acts as a transit point
> for that client.  Additionally, a client may have multiple transit
> points to various border routers wich may or may not include the RR. My
> point is if you only run two RR's for this 35 router network, you're
> only going to get two best-routes for any given NLRI (assuming
> client-client-reflect and no intra-cluster meshing). If you have 3+
> potential exit points for a given NLRI, you run into inefficiencies. 

I was assuming a set of pops tied together by a set of transit circuits.
In which case, the top level routers would act as the RR-servers and exit
points out of the pop. In the completely flat network scenario, the above
point is valid. In the hierarchical network, the exit routers will make a
decision that the pop routers would have made for NLRI coming in over the
transit mesh.  The only potential difference is if the same NLRI is also
visible from a router in the pop.


> > As far as ibgp meshes go, 75 routers in a full mesh is trivial with the
> > caveat that routers dave cooper is running tend to be on the high end of
> > the scale as far as horsepower goes. 

> > > I would not suggest running confederations.  
> > 
> > Reasons?
> 
> On 35 routers? Let me recant... are their reasons he *should* run confeds?

None whatsoever.  I thought you were saying RR's were a better choice over
confederations, not just in this scenario.

/vijay


-
Send 'unsubscribe' in the body to 'list-request@inet-access.net' to leave.
Eat sushi frequently.   inet@inet-access.net is the human contact address.