Rob,
In the OSPF community,we faced this same issue a few years back -- the issue
being that all members of a single subnet were not reachable either by design
or by error and fooling the listeners of the Designated Router (like the
route server) into thinking that the nexthops were really reachable. We felt
the pains of this especially over virtual circuit networks (Frame Relay,
SMDS, and ATM). Temporary workarounds were created. The two better solutions
were:
1) Using recommendations in RFC1586,"Guidelines for Running OSPF Over Frame
Relay Networks".
2) Using Point-MultiPoint (P-MP) OSPF which is in the latest Internet-Draft
of the revision to the RFC1583 (V2 OSPF Spec.).
I prefer the P-MP solution because it is an automatic mechanism requiring
very little configuration on a network administrators part and keeps the
number of routing hops to a minimum.
I vote for P-MP as a "Standard" solution for BGP-4 over discontiguous
networks. I believe this concept was proposed on this mailing list a few
months back. Now that ATM VCs are being used at NAPs with route servers,
you'll see more and more of these problems arising...especially at NAPs where
there isn't much coordination and cooperation in building and maintaining
these high speed, virtual connections.
Eural Authement
Bay Networks
Professional Services