Next, look at what needs to be replicated.
- Do you need to replicate the whole database or just certain schemas?
- Do only certain tables need to be replicated?
- Of those tables are there only specific rows or columns that should be replicated?
- What changes should be replicated?
- Just DML?
- Which DML; inserts, updates, and/or deletes?
- Do you need to replicate object structure changes, aka DDL, as well?
- And again, what DDL changes need to be distributed?
- Do you need to replicate dependant application objects; like packages, indexes, views, privileges, triggers, and so on?
- What are the dependencies between these objects?
- Many of these objects require special handling (privileges, procedures, views, sequences, and triggers) which means additional configuration.
Also look at what role each site will play. Will the site be:
- A master: Changes are made at the site and sent to other sites. These sites usually contain the full copy of the distributed data and communicate continuously with the other master sites. Any interval between Propagation should be relatively short to avoid data divergence.
- A secondary or slave: The site contains all, or a subset of, the distributed data and receives changes from one (ideally) or more masters. Direct changes are rarely made at this site. If changes are made at this site, any conflicts with master site data will result in the change being overwritten by the master site data at the next Propagation interval.
- A directed network site: A master or secondary site that receives changes from another site, may or may not apply the change to itself, then passes the change to another database.
A word of caution on how many sites you choose to be master sites. For each site that you allow to be a master site, you increase your risk of data divergence exponentially. The more master sites you allow, the more conflict resolution definitions you will need to put into place. As the number of master sites increases, conflict resolution becomes statistically impossible, and over all maintenance and administration becomes unfeasible.
Understanding what you want to replicate will help you determine how much configuration work you will have ahead of you. It will also help you identify potential problem areas that may require special handling.