Whenever I'm describing the idea of running BGP across parallel serial links with duplicate IP addresses (like I did in the November IP Corner article, Load Balancing in BGP Networks, section External BGP Load Balancing), there's always someone asking “does it really work?” … so I'm enclosing a tested working configuration.
| AS 11 | AS 12 |
|---|---|
interface Serial1/1 | interface Serial1/1 |
AS11#show ip bgp summary ¦ begin Neighbor… then the BGP routing table …
Neighbor V AS MsgRcvd MsgSent TblVer InQ OutQ Up/Down State/PfxRcd
10.0.1.10 4 12 13 12 2 0 0 00:09:02 1
R2#show ip bgp | begin Network… and finally the internal details of the CEF entry (that's the only way to actually verify that the load balancing is taking place):
Network Next Hop Metric LocPrf Weight Path
*> 172.16.0.0 10.0.1.10 0 0 12 i
AS11#show ip cef 172.16.0.0 internal
172.16.0.0/16, version 35, epoch 0, per-destination sharing
0 packets, 0 bytes
tag information from 10.0.1.10/32, shared
local tag: 17
via 10.0.1.10, 0 dependencies, recursive
next hop 10.0.1.10, Serial1/1 via 10.0.1.10/32
valid adjacency
tag rewrite with Se1/1, point2point, tags imposed: {}
Recursive load sharing using 10.0.1.10/32
Load distribution: 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 (refcount 2)
Hash OK Interface Address Packets Tags imposed
1 Y Serial1/1 point2point 0 none
2 Y Serial1/2 point2point 0 none
3 Y Serial1/1 point2point 0 none
4 Y Serial1/2 point2point 0 none
5 Y Serial1/1 point2point 0 none
6 Y Serial1/2 point2point 0 none
7 Y Serial1/1 point2point 0 none
8 Y Serial1/2 point2point 0 none
9 Y Serial1/1 point2point 0 none
10 Y Serial1/2 point2point 0 none
11 Y Serial1/1 point2point 0 none
12 Y Serial1/2 point2point 0 none
13 Y Serial1/1 point2point 0 none
14 Y Serial1/2 point2point 0 none
15 Y Serial1/1 point2point 0 none
16 Y Serial1/2 point2point 0 none
Posts
why on earth wouldn't you use ebgp multi-hop?
ReplyDeleteBecause:
ReplyDelete(A) it poses a slight security risk (although that can be reduced these days).
(B) you need to configure the routing toward the EBGP next-hop.
Neither one of these is a big issue, but there's no reason you shouldn't use this feature if it's available.
I wonder if this is widely accepted in the ISP world. For example, if one of the circuits is behaving erratically and need to perform a series of ping packets to confirm, how can I control the ping to travel onto the same circuit and back?
ReplyDeleteHonestly this is a neat trick I must admit, but configuing it would require you to disable one serial interface before you can configure the same IP address on another serial interface or else IOS would complain that you have overlapping IP addresses on the box.
@william: The issues you raise are probably the true reasons this trick is not widely used.
ReplyDeleteAs for configuring it, IOS allows overlapping addresses on WAN links (that's the essence of this design).
Maybe I'm missing something, but... what's wrong with MLPPP for this?
ReplyDeleteMLPPP is perfect if you can use it. You might have performance/hardware issues (although that should not be too much of a problem lately) or your WAN encapsulation might be something else than PPP.
ReplyDeleteI might be missing something, but how can you put the same IP address on two interfaces on the same router/vrf? Doesn't the router complain about duplicate IP addresses?
ReplyDeleteThe "trick" is that the router does not complain about overlapping addresses if both interfaces are WAN links (could be even multi-access links like FR/X.25/ATM).
ReplyDeleteHopefully not too stupid of a comment. When I lab this using physical interface, frame relay encapsulation, and frame relay maps cef does not load balance.
ReplyDeleteHowever, when I create point-to-point subinterfaces it works identically to what you have described using frame relay encapsulation.
It really is dependent on the interface being flagged as point-to-point :)
can i use 2 AS number on single BGP running router
ReplyDeleteNo, but you can pretend to be a different AS to a neighbor:
ReplyDeletehttp://wiki.nil.com/BGP_Local-AS_feature