Re: EIGRP best path selection with multiple successors

Just to expand a bit - Let's say you have a simplified version of the DMVPN
network in this documentation. Say you have a single hub and 3 spokes, S1,
S2 and S3 and your network is a DMVPN phase 2 design running EIGRP. We
know that in Phase 2 design on the hub, we would both disable split-horizon
and we would use "no next-hop-self eigrp" on the hub. This allows the DMVPN
hub to learn spoke routes on the tunnel interface, then turn around and
advertise them back out the same multipoint tunnel to other spokes with the
original next-hop intact. Thus, we can allow dynamic spoke to spoke
tunnels.

Now, say S1 and S2 are both connected to the same LAN segment and both
advertise this subnet with the same composite metric to the hub. The hub
will have 2 successors to this network, but as the article states, by
default will only advertise one of them to S3. The question is which one
and why? The article only mentions that EIGRP will select one but does not
go into detail as to how this happens. At S3, is the next-hop of the LAN
subnet going to be S1 or S2 and why?

On Wed, Dec 11, 2013 at 5:24 PM, Joe Astorino <joeastorino1982_at_gmail.com>wrote:

> So, the new EIGRP "add-path" feature is intriguing to me as documented
> here
> http://www.cisco.com/en/US/docs/ios-xml/ios/iproute_eigrp/configuration/xe-3s/ire-add-path.html
>
> As we know, EIGRP can indeed have multiple equal cost paths. In that
> case, we will see more than one successor in the EIGRP topology table.
> However, only one is actually sent to EIGRP neighbors by default, as the
> article points out.
>
> In certain DMVPN situations, this does not allow spoke to spoke load
> balancing, hence the point of the add-path feature as explained.
>
> All that is well and good, but I couldn't help asking myself as I was
> reading "wait a minute, by default without all this, which successor would
> it send?"
>
> I have spent some time googling around and such and have labbed some
> things up, but have not got a definitive answer. Anybody know?
>
> In my lab test I had frame-relay hub and spoke network setup. The spokes
> are both connected to the same ethernet segment and I had them redistribute
> that ethernet segment into EIGRP with the same metric, but using different
> route tags. My initial experiments seemed to indicate that the successor
> sourced from the neighbor with the lowest IP address were sent upstream to
> other EIGRP neighbors.
>
> --
> Regards,
>
> Joe Astorino
> CCIE #24347
> http://astorinonetworks.com
>
> "He not busy being born is busy dying" - Dylan
>

-- 
Regards,
Joe Astorino
CCIE #24347
http://astorinonetworks.com
"He not busy being born is busy dying" - Dylan
Blogs and organic groups at http://www.ccie.net