After servers have been configured for replication and the source and destination data tables have been synchronized, a replication process (the log reader) monitors the transaction log of each database set up for replication. When a transaction occurs on a published table, it is marked for replication. When the log reader finds transactions marked for replication, it applies them to the distribution database. The transactions are held in the distribution database until they can be distributed to the appropriate destination servers and applied to the destination databases.
Only committed transactions are sent to destination servers. Replicated changes to a table are guaranteed to be sent to the destination servers in the order in which they are committed. This ensures the subscribers receive transactions in the same order in which they were applied at the publisher. Because this is a log-based solution, maximum concurrent access to data is maintained. Replication does not exclude user access to destination tables during the associated inserts, deletes, and updates.
During normal operation, transactions marked for replication are preserved in a publication database's transaction log until they have been moved into the distribution database by the log reader. After this occurs, the transaction log of the publication database can be truncated using normal procedures. Records still waiting to be replicated are not truncated. (The transaction log on the publisher can be dumped without interfering with replication, since only transactions not marked for replication are purged.)