First I would like to that Michael Kaplan and Jack
MacDonald for the information they have bestowed upon me
directly or indirectly (no pun intended) thorough others
postings on the newsgroup.  My understanding is still
incomplete as I approach my first attempt at doing this in
the real world.  I was hoping to get a few more thoughts
on my proposed conversion to a farm topology beform
implementing it (measure twice and cut once).
Current setup:
1. A design master sitting on a server on a wan
2. A replica on each of the client pcs on the lan (why are
there replicas on each client on the lan? ... because they
substantially improved the performance for folks on the
lan)
3. A replica on each of the client pcs which access the
server via wan.
4. In code each of the replicas is synchronized directly
with the design master via JRO upon startup and shutdown
of each replica.

Problems with the currect design ..
1. occasionally the design master becomes incommunicable
for syncs and some intervention needs to be taked to
repair it.
2. I gather the setup violates some established principles
for good design of a replication stategy

Proposed Design:
1. A farm consisting of three replicas all sitting on a
share on a spare machine on the lan.  RepMan would manage
this set.
2. A replica would reside on each user machine whether
they are on the lan or on a wan connection.  RepMan would
also be on each machine and manage the local copy.
3. Indirect synchronizations would be initiated from the
replicas via their local synchronizer with the farm.
4. The design master would stay on the server and be
unmanaged.

Proposed conversion process:
1. Create a share on a spare machine on the lan
2. Install rep man on the machine.
3. create three replicas on this machine and manage them
via repman.
4. Install repman on each machine with a replica.
5. Remove JRO code which currently causes direct syncs
with the design master.
6. Use the client copies of repman to establish a schedule
for indirect syncs of the local managed replica with the
farm.

Questions:
1. It this reasonable?
2. The local replicas on the client machines will not be
visible to the farm or the rest of the network.  However,
since the syncs will be initiated from the client machines
is this necessary?  It seems that the drop box might need
to be visible .. but could this reside on a network server?
3. This is a more fundamental question ... I am still not
sure I understand why indirect syncs are important.  I
understand how the farm can repair itself and bring
members that have been labels unavailabe for syncs back
into the loop.  Is it that indirect synchronizations
minimize the problems that the self repairing properties
of the farm correct?

Thanks

JD