The hub/spoke topology you mention makes sense to me. The hub functions as a distribution server to the spokes in this arrangement.

At a database level, replication of DBs is subject to the same ACL settings as any other type of access, so be sure your databases have the proper ACL settings.

There are also selective replication settings, but they're not automatically added.

Servers in the same NNN do not need connection documents for mail routing, but do need connections documents for replication.

I've struggled with actual configurations before as well, but lately I try to set both to and from connections documents, and usually try to stagger the times; ie Server 1 to Server 2 every 30 minutes starting at 00:00, and then a second connection document from Server 2 to Server 1 every 30 minutes starting at 00:15.

That way if a replicator goes down or gets hung on a long replication, replication still happens from the other server. and the replication 'load' is spread out instead of all servers replicating at the exact same time.

Also, if your hardware is adequate, be aware that you can set up multiple instances of replication - on your hub server, for example, I might recommend adding 'Replicators=2' to your Notes.ini.

If you have servers in clusters, you'll probably want to set up replication between the clustered servers as well as the hub.

Thank you guys, it is always good to hear someone else opinion and how your setup looks like...

This is good tip, connection documents in both directions but with different time interval!

I've setup replication interval to 10 minutes and adminp (interval) to 15 minutes so I'm always sure that both replication and adminp will be always performed in max of 15 minutes.

Using connection documents for replication between all servers is also a method to ensure that the replication occurs (as a backup).  If one replication fails , then the next server can pick up the failed sequence and ensure that all documents are up to date.