AD across multiple campuses

[Jeff Kell <jeff-kell () UTC EDU>]

We are again revisiting a "single forest" integration of AD across
multiple campuses in the system, and the inevitable pros/cons and issues
are starting to roll in.  I'm interested in any information / pointers /
feedback / suggestions from other sites that have done or tried this
approach.  Specifically with regard to routing/addressing and overall
security (not just the AD components)...

Each campus is essentially a separate network entity now, no private
links / leased lines / hocus-pocus.  We have our individual internet[2]
providers.

Most sites are using RFC-1918 addresses internally, to varying degrees,
some are required to (lack of public IPs).  The RFC-1918 space currently
overlaps.

Each site will have it's own domain (tree?) but there is a desire for
cross-site replication and redundancy of domain controllers.

The current "desire" from the design group (I'm certainly no AD guru,
I'm network/security) states that each user on each campus must have a
globally unique IP address, such that "any" user at "any" campus can
authenticate to "any" of the domain controllers.  Some us will have to
renumber our 1918 space, but it is doable.  Private links (either
leased, site-to-site VPN, or MPLS-provider private vlans) have been
proposed to route the AD client/domain traffic within the system.

Having survived Blaster, Slammer, Nachi, Conficker (so far), and other
network nightmares exploiting Microsoft infrastructure in the past, I'm
very concerned about having one big flat system-wide routed domain.

Restricting the "private" links to "only" AD-related traffic seems
troublesome as well (firewalls, policy routing, etc), but perhaps not
insurmountable.  There is a great deal of existing inter-campus traffic
(operating through traditional NAT/firewalls) for centralized SAP/R3,
video traffic, distance learning, distributed printing, and other
cases.  Much of this is freely intermixed on networks that would also
have AD clients/servers.

Is all of this really necessary to provide the desired result?  Is
anyone doing this over "the public internet" ?

Can DC-to-DC replication happen across NAT, without all this other
infrastructure/exposure?

Can client-to DC authentication happen across NAT?

Is there some way to keep the existing site network/routing autonomy?

Jeff Kell
University of Tennessee at Chattanooga