BGP convergence

[chinkevi_2]

hello,

is there a way for BGP to switch and prefer IGP, after the routing table uses BGP as prefered route?

here is a scenario:

- 2 different sites connect to an ISP, so both CE has different AS numbers.

- both CEs are advertises each site's prefixes with network statement.

- both sites reach each other via ISP cloud as normal.

- issue arise when both sites connected to each other via backdoor link, and run IGP - ospf or eigrp. So both routers are doing IGP and BGP.

- routing table of both CE still use BGP as preferred routes, until do clear bgp neighbor.

regards

[milan.kulik]

Hi,

what about BGP

network backdoor

command?

See http://www.cisco.com/en/US/docs/ios/iproute/command/reference/irp_bgp4.html#wp1012431

for details.

BR,

Milan

[jcozzupoli]

hello,

have you tried doing prefix filtering and metric manipulation using route-maps, prefix-lists and even redistribution between IGP and BGP? You could also use IPSLA and tracking of routes/prefixes

Also, is this "backdoor link" always active? What I mean is, is it a backup link between the 2 CEs or is it the primary path? I'm guessing its a backup link. If its backup, tracking a destination network is a good way to force routing to failover if done correctly.

Regards,

Joe Cozzupoli.

[chinkevi_2]

thanks for the replies.

the backdoor link is the primary between the sites.

Both CE have network statement for both site's prefixes, so to force the prefer path that is locally originated.

"network x.x.x.x backdoor" would not be needed here due to both CE are advertising the same prefixes.

the traffic pattern should be such that all incoming traffic is via CE1.

CE2 has as-path prepend to make the route less preffered for incoming traffic.

so the question is why bgp did not prefer igp when the backdoor link was connected, and require bgp reset?

we are thinking this may recur when backdoor link is broken and recovered again.

[jcozzupoli]

Hi,

Well to start with eBGP routes will be prefered over IGP routes. But i'm sure you know this already.

Secondly, are you redistributing the IGP routes into BGP? If so, have you tweaked any metrics to make sure LAN routes (CE to CE via Backdoor link) are given different metrics to make them prefered over the eBGP path?

[chinkevi_2]

i think it is more like bgp do not update igp learnt routes for some reason, after the backdoor link is connected.

to answer your question, i would think the behavior should be consistence but it is not. The lan uses igp only and metric is set to prefer a CE as primary default.

i have attached debug for reference from a test.

10.10.10.0/24 is CE1 lan address

20.20.20.0/24 is CE2 lan address

30.30.30.0/24 is backdoor link behind the 2 CE that is 1 hop away.

the debug is taken from CE2. Only 10.10.10.0/24 do not get updated. Others updated and put local source routes (network statement) to bgp table.

the network statement on both routers are the same.

[chinkevi_2]

i think i figured out why igp routes is not updated in bgp table.

bgp network statement is to make the prefix locally source, with condition that the prefix must be in routing table.

As the prefix is already in routing table as BGP learnt route already due to ebgp preffered over igp, so i guess there is a rule say the prefix cannot be locally source if the prefix is bgp learnt route in routing table.

when clear the route entry, enable igp learnt route enter the routing table first, so bgp table is eventually updated the prefix as ebgp and local originate, and locally originated route becomes preffered.

let me know if this is correct.

my next question is.. if i need to redistribute igp like ospf due to large number prefixes, so i cannot do bgp backdoor, any other option to get around this?