In this study, we configured and tested the TimesTen cache in asynchronous write-through (AWT) mode. In this mode, the propagation of transactions from the TimesTen cache to the backend Oracle Database is asynchronous and is done in batches. When the application completes a transaction and issues a commit, the commit executes and completes in TimesTen and the transaction is then protected by TimesTen's native persistence mechanisms.
When the transaction commits in TimesTen, it becomes a candidate for propagation to the backend database. The TimesTen replication mechanism will capture the transaction from the TimesTen log stream (usually from the in-memory log buffer) and add it to the current batch. When sufficient transactions have accumulated (approximately 256 KB of data), the batch is applied to the backend database and committed there.
The propagation and the commit-apply process are parallelized and use dependency tracking to maintain the correct ordering in terms of both DML operations and commits. The replication propagation is real-time and the delay between a transaction committing in TimesTen and committing in the backend database is typically a few milliseconds.
Note: TimesTen cache also supports synchronous write-through (SWT) mode, which provides a tighter synchronization between the TimesTen cache and the Oracle database but at a considerable cost in terms of DML performance. Most customers prefer to use AWT to maximize performance. If the TimesTen cache is implemented as a HA pair using TimesTen replication (recommended for production configurations), then updates originating in the cache are fully protected against loss due to failures.