Re: iBGP Table Entry Qualification

kfarrington · ‎09-30-2003

Hello all,

I have a question re the setup below.

I have two routers peering eBGP with an external AS 350 and iBGP between the routers which (just for info) are separate locations.

Now if I receive a route (say 168.1.0.0) that has a metric inbound on both routers (router76 route for this network has metric of 33 ,and Router12 has metric for this route of 42) it looks as if the router that has the lowest metric from the route, does not accept the iBGP route from the other router with the higher metric, but the router with the higher metric accepts the iBGP route from the router with the lower metric.

If I receive a route where the metric is not set and all attributes are the same, it send each other the iBGP route (like 168.41.0.0) and uses it's local exit point via eBGP to that network.

Please can someone confirm that this is normal operation for iBGP.

Many thx indeed,

Ken

-------------

output below - if it can be interpreted :)

-------------

on Router76

router bgp 300

neighbor 150.5.36.129 remote-as 300

neighbor 192.168.42.254 remote-as 350

Router76>sh ip int brief

Interface IP-Address OK? Method Status Protocol

Ethernet0 unassigned YES unset administratively down down

Ethernet1 192.168.42.253 YES NVRAM up up

Loopback0 150.5.32.65 YES NVRAM up up

Network Next Hop Metric LocPrf Weight Path

-------------------------------------------------------------------------------------------------------

*> 168.1.0.0 192.168.42.254 33 0 350 i This entry only appears once

-------------------------------------------------------------------------------------------------------

*>i168.224.0.0 192.168.41.254 31 100 0 350 i

* 192.168.42.254 42 0 350 i

-------------------------------------------------------------------------------------------------------

* i168.41.0.0 192.168.41.254 100 0 350 i This entry appears on both

*> 192.168.42.254 0 350 i This entry appears on both

-------------------------------------------------------------------------------------------------------

On Router12

!

router bgp 300

neighbor 150.5.32.65 remote-as 300

neighbor 192.254.41.254 remote-as 350

!

Router12>sh ip int brief

Interface IP-Address OK? Method Status Protocol

Ethernet0 unassigned YES unset administratively down down

Ethernet1 192.168.41.253 YES NVRAM up up

Loopback0 150.5.36.129 YES NVRAM up up

Network Next Hop Metric LocPrf Weight Path

* 168.1.0.0 192.168.41.254 42 0 350 i

*>i 192.168.42.254 33 100 0 350 i

-------------------------------------------------------------------------------------------------------

*> 168.224.0.0 192.168.41.254 31 0 350 i This entry only appears once

-------------------------------------------------------------------------------------------------------

*> 168.41.0.0 192.168.41.254 0 350 i This entry appears on both

* i 192.168.42.254 100 0 350 i This entry appears on both

-------------------------------------------------------------------------------------------------------

ruwhite · ‎09-30-2003

This is correct. iBGP, like all distance vector protocols, will split horizon the information it advertises. So, if Router76 is actually using the path through Router12 to reach a given destination, it will not advertise that destination to Router12. However, in the case where both routers are preferring their eBGP path to reach the destination, they will both advertise the route to each other (since they aren't using each other as their next hop).

Hope this helps...

Russ.W

kfarrington · ‎09-30-2003

thx, kinda gottit

you mean that if R76 is using an iBGP path through R12 for this network, R76 wont advertise its route (any iBGP/eBGP route to that network) to R12 for R12 to enter into its table

Can I just confirm the semantics on this please Russ,

many thx indeed

ruwhite · ‎09-30-2003

Correct. Of course, R76 won't advertise iBGP learned routes to R12 anyway, as Alvaro noted below. It only shows up here because these two have a route they are each learning from an eBGP peer, and advertising to an iBGP peer.

Russ.W

aretana · ‎09-30-2003

Ken:

In short, only the best path is propagated...and no iBGP learned paths are advertised to iBGP neighbors.

For the case of 168.1.0.0...the entry with the 192.168.42.254 next-hop is prefered by both routers. for 168.41.0.0, each router uses it's own exit, so the path is in both places.

Note that BGP doesn't enforce split horizon strictly; in some cases a path may be sent to the originator as a matter of routing operation -- the receiver in this case should discard the route anyway. This is the specific case of peer-groups, for example.

Alvaro.