Cisco Support Community
Showing results for 
Search instead for 
Did you mean: 
New Member

BGP Mulipath load share

i have a customer connected to my network with 15 link, how can i run BGP with him to load share the traffic over the 15 links



Re: BGP Mulipath load share


AFAIK, it's possible to configure BGP load-sharing up to 6 lines. See

Wouldn't it be possible to run multilink PPP over some of your lines?



New Member

Re: BGP Mulipath load share

Sir, i know it is possible to load share BGP up to 6 lines..but i have 15 one !!

if i replaced the 15 ports with an E3 port and a MUX for physically grouping all of the interfaces, will this be usefull and BGP will treat it as only one Path and a MUX will distribute traffic evenly.???


New Member

Re: BGP Mulipath load share

what about if i configured two EBGP sessions with EBGP multihop (for load sharing on the 6 pathes)where each BGP session is established based on a different Loopback interfaces, i.e. Session one between loopback 1 on my router and loopback 1 on customer router, while session 2 between loopback 2 on my router and loopback 2 on customer router.

will this load share accross the 12 pathes ??

any one can give me help..??

New Member

Re: BGP Mulipath load share

I would not expect it to work since it is 6 paths per route, not six per peer. If you and your customer are exchanging atleast 3 routes each, you could break the routes into 3 groups, and using BGP attributes (e.g. MED/AS-PATH/local pref) make group1 prefer links 1-5, group2 links 6-10 and group3 links 11-15.




Re: BGP Mulipath load share

You can load shar on up to 8 paths. You can't run two peering relationships between the same pair of routers, so that option is out. You could bundle the links using mlpp....

Let's see, you could try use route maps to set the next hop on some percentage of the routes to a different link outbound, but I don't know if that will work. You could also try and build up the peering relationship between two loopbacks, and then build out recursion to cover all 15 paths, but I don't know if CEF will handle the recursion correctly in load sharing in this way, and the hardware probably won't, either.

I'd say your best option is to bundle the links.



Re: BGP Mulipath load share

If you're a bleeding-edge type of person you might consider using 12.3(2)T, which allows load sharing over 16 paths rather than just 6.

Note however that it hints at platform restrictions for support of this command, yet doesn't list which platforms it is/isn't supported on. Feature Navigator may provide more specifics.


Re: BGP Mulipath load share

There are platform restrictions on this--I'm not certain why they didn't list them here, it could be because it's a confusing maze of different restrictions for different platforms, and they generally let the platform people sort it out, and post it for each specific platform.

So, the limitations you are going to hit, in general, are based on the number of equal cost paths a secific line card, with a specific hardware implementation, can handle.

Some information is here:

For the 12000, it's going to be based on the engine type of the line card, so you'd have to look at each line card and figure out what they are limited to in hardware for load sharing. I don't see a URL for information about which line card will do how many....

Here, it says the 6500 can do at least 8:

But I don't see anything about that being a hardware limitation, or they will be able to increase it in some future release. I assume there is a hardware limitation on this box, since it's an ASIC that does the switching.

On all software based platforms, you should be able to switch across all 16 links, running this code, which would include the 7200, 3600, and 2650.


CreatePlease to create content