Redundancy for recursive resolvers w/dual ISPs

[Chris Timmons]

We have a residence hall network and an administrative network, each
running on discrete address space routed by different ISPs.  Each network
has separate namespace and operates a pair of its own recursive name
servers (bind8) to handle client queries.  Each side has a split-horizon
DNS configuration with externally visible DNS handled by a separate
instance of named; for this discussion I'm only talking about the
internal, recursive name servers which process requests for client
resolvers.

Due to historical factors and address space transitions, the name servers
servicing the residence halls are running with IP addresses from the admin
network.  If the ISP for the admin network has an outage, it effectively
creates an outage for the residence halls because their resolvers go dead
in the water.

In addition to making the long procrastinated transition to bind9, I am
considering how best to resolve this architectural flaw in our DNS
topology.  Obviously I will configure the residence hall name servers with
IP address space routed by their ISP, which solves the main problem.

My question is, should I consider a scheme which would allow the name
servers from both networks to share a site-wide cache?  I thought of a way
to do this by configuring two new site-wide recursive name servers, one
from each network - and then configuring each existing name server to
forward to the pair.  The good news would be the site-wide cache and
potentially the small benefit (stop applications from hanging on DNS and
let them get right to the absence of L3 connectivity :) if 1 of the 2 ISPs
goes down.  Unfortunately such a configuration seems inefficient and
overly complex.  Should I bother?  Any magic bind9 features to let the
caches corrupt each other without all of that forwarding?

-Chris Timmons, Network Engineer
 Central Washington University