I set up a test replica farm today based on what I read on Michael
Kaplan's Trigeminal site (www.trigeminal.com).

I created three replica's for the farm and have them all managed by
the same sychronizer.  They are set up to replicate every 15 minutes
which seems to be the shortest interval allowed by Replication
Manager.  The hub is located in a directory called Hub on a server and
the farm relicas are named Farm Replica 1, Farm Replica 2 and Farm
Replica 3.  I also set up a replica on another machine called Remote
Replica 1.  This replica is managed by its own synchonizer.  (Question
- can more than one synchronizer run on a given machine?).  Remote
Replica 1 was set up to replicate with the sychronizer managing the
farm every hour.  I added a record to Remote Replica 1 and then
watched to see what happened.  When the time came for the
sychronization of Remote Replica 1 and the farm synchronizer came and
passed I observed the following.  Farm Relica 1 contained the new
record while the other two farm relicas did not.  So apparently the
farm synchronizer selected farm relica 1 for the sychronization with
Remote Replica 1.  I continued to wait to see what would happen when
the farm synchronizer synchronized its internal members.  When the
time came and passed I observed that now Farm Replica 2 also had the
new record but that Farm Replica 3 still did not.  Upon reviewing the
synchronization log what appeared to have happened during the
sychronization of the farm synchronizer's managed replicas was as
follows:
Farm Replica 2 synchronized with Farm Replica 3
Farm Replica 2 synchronized with Farm Replica 1

Thus, the new record was not transferred to Farm Replica 3 during this
process.  During the next synchronization of the farm synchronizer's
managed replicas the added record will be propagated to Farm Replica
3.  So .. bottom line is that it appears that there can be some lag in
the propagation of updated or new data among farm members .. not a
problem .. just an observation. This lag can be substantially longer
for remote replicas since they would presumably be sychronizing with
the farm less frequently.

Another observation - The managed replicas of a given synchronizer
synchronize with each other in a direct manner.  There doesn't appear
to be an option to have them synchronize in a indirect fashion.  This
may seem like a silly observation to those who have some substantial
experience in this area but the reason it stood out to me was as
follows.  I thought one of the advantages of indirect synchronizations
was that they can minimize corruption problems.  Further, I thought
the likelyhood of corruption problems was related to multiple
databases being open and communicating with each other at the same
time as opposed to passing notes in drop boxes.  Thus, the question
that I have is ... with multiple replicas in a farm doing frequent
direct synchronizations with each other ... isn't there an increased
chance for corruption?

Yet another observation - I installed replication manager on my
machine.  I confurged the synchronizer to manage the replica farm
which was residing on the server.  Bottom line is that I don't need to
install Replication Manager on the server ( which in my opinion is a
good thing since the less I have to rely on a business IT folks to get
things done the better (not that I have anything against IT folks in
general you understand :)

One other question related to problems that can occur with farm
replicas and how they are corrected.  What types of problems can the
internal direct synchronizations fix automatically and what types of
problems can't they fix.  What types of data loss can take place?  Is
it possible to set up the system so that no data loss will take place
or at least so it's obvious when data loss took place?

What type of regular maintainence is required with manageing the farm?

Thanks in advance for your thoughts - JDK