How does iBGP multipath load sharing work?

[weizixi]

the iBGP multipath loadsharing feature can enable load sharing at multi path,but if that the paths are in different bandwidth, for example one is 10M and another is 100, should the router divide the traffic into 1:10 then load sharing?

[thisisshanky]

BGP uses its own algorithm, to find out the best path ( based on BGP attributes). It doesnt care about bandwidth on the interface. So if two routes are installed via the 10 M and 100 M path, first the routing table is scanned, to find a route. The router sees that there are two routes, via the 10 M and 100 M path, both as equal cost route. Now it sees what type of switcihng (fast switching, or CEF, or per packet load balancing). Depending on that switching will take place. So you really cannot rely on traffic to be shared on a 1:10 basis.

But reading another document, i find that, if you conifugre "maximum-paths eigbp" which enables maximum path load sharing on the router, for both EBGP and IBGP. The default loadbalancing technique used, when this command is executed is Unequal cost load balancing. Since Unequal cost loadbalancing is the default, I think, it should do a 1:10 load sharing of packets. Check this command out.

http://www.cisco.com/univercd/cc/td/doc/product/software/ios122/122newft/122t/122t4/fteibmpl.htm

[ruwhite]

If the local pref, med, as path length, and igp metric are all the same, iBGP will install both paths in the routing table, which then allows CEF (or some other switching path) to load share between the two paths. BGP generally will not take into account the bandwidths of the links across which the paths are being learned.

However, there is a feature, dmz link bandwidth, which allows bgp to do unequal cost loas sharing based on the bandwidth of the exit point link (the bandwidth of the link across which the ebgp session is built). The bgp speaker which learns the route through ebgp will put a special community on the routes as they are learned (called the dmz bandwidth community) which indicates the bandwidth of the exit link.

An iBGP speaker which then chooses to load share between two links will actually use this dmz bandwidth community to change the round robin counters in the rib, causing the two paths to be unequal cost load shared based on the exit link's bandwidth.

Take a look here:

http://www.cisco.com/en/US/partner/products/sw/iosswrel/ps1829/products_feature_guide09186a008012db46.html

for information on this feature.

Russ.W