> IBGP is the IGP. Each confederation is made up of
> fully-meshed IBGP speakers in a peer-group. Intra-sub-as
> routing is not a problem, it is routing between confederations
> that is posing the trickiness.
Rather than creating a fully-meshed sub-AS configuration with only iBGP peering between the sub-ASs, I believe a more common/optimal model for deploying BGP confederations with fully-meshed sub-ASs is one in which requires that all sub-ASs share a common IGP (not iBGP, rather RIP, OSPF, ISIS, EIGRP, etc..) and allow the NEXT_HOP value to be set by the first router in the confederation, etc... If you do manage to get things right without a true IGP, most would still benefit greatly when considering convergence across the network.
If containing IGP (not iBGP) scaling is the goal and you'd prefer to use different IGPs within a sub-AS, you'll likely want to go with a centralized confederation design whereby all the sub-ASs exchange destination information via a core or central sub-AS. Of course, this has a few obvious offshoots as well. And if that big of a deal, you could always use eBGP and the private-AS stuff on your Internet-connected BGP routers.
I do suppose an implementation could add knobs to support policy definitions based on AS_CONFED_SEQUENCE/AS_SUBCONFED_SET values. However, it seems there are plenty of more optimal options available that would allow one to avoid doing this.
-danny
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:06 EDT