The Advanced Replication synchronization technology relies on a master/replica model. A single computer keeps a master or official copy of the data. All other servers (replica databases) in the system can also have a copy of some or all of the master data. When the replica database changes data, that data is unofficial until it has been sent to, and accepted by, the master database. After the data has been accepted on the master database, the replica database that sent that data (and other replica databases as well) can request a copy of the new official data. In this way, replica databases can become temporarily out of sync with the master database, but such differences can be corrected quickly.

The Advanced Replication architecture supports a virtually unlimited number of replica databases to have access to data. Each replica database can read and write the data, as long as users know that local writes must be accepted at the master database before they are official. Replica databases do not need to be connected to the network all of the time. A replica database can be stored on a mobile device or computer that is only connected to the network part of the time. This allows replica databases to operate for long periods independently of the master database, providing flexibility. It also means that the entire system is not disabled just because the master database server is shut down for maintenance or to correct a problem. Distributed systems based on the Advanced Replication technology are inherently robust with respect to isolated failures.

A replica database can hold not only copies of some or all master data, but also local data that is not shared with the master database. A single computer can hold a mix of local and shared data.

The master database stores the data and the synchronization definitions. The synchronization definition data includes catalog, database schema, publication definitions, registrations, subscriptions, user access definitions of the replica databases, and so on.

The replica database stores local data and transactions that are to be propagated to the master database. All replica data, or a suitable part of it, can be refreshed from the master database whenever needed by sending a REFRESH statement to one or more publications. The local data includes data that is typically a subset of the master database, as well as system tables that contain information specific to the particular database.

In a multi-tier synchronization environment, synchronized databases can be configured to serve a dual role as both a master database and a replica database. These roles are established by creating a catalog and defining it to be both a replica database and a master database.