...

The IGP rerouting with modern link-state IGPs is sub-second so the redundancy of interior paths is easy to achieve (perticularly if you use tricks like BGP confederations which eliminate need to recompute iBGP routing in case of IGP changes).

Can you be more specific about this? How do you avoid the recompute and also avoid persistent routing loops after a topology change?

When BGP confederations are used, iBGP routes point to the exit interface address in a satellite AS, instead of the border router's address. That address is invariant (unlike the border router's, which may change with changes in interior routing). This allows to preserve BGP table unchanged in case of IGP changes, and simply re-run one step of recursion (assembly of iBGP and IGP tables) instead of sending costly BGP updates with new border router's addresses. Of course, this works only if backbone is not fractured (i.e. threre's still connectivity to all exits). As for routing loops we all know whom to thank for the wise decision to give packet forwarding higher priority than to handling routing updates, and for using textual regexp matching in policy evaluation, too. --vadim