Synchronous and Asynchronous Replication of Cache Objects When you set a backup count, you define the number of backup object copies to make, in addition to the primary cache object. Backup cache writes can be done synchronously or asynchronously. There is a difference between the behavior of TIBCO BusinessEvents DataGrid and for Oracle Coherence as regards replication. TIBCO BusinessEvents DataGrid Uses Asynchronous Replication TIBCO BusinessEvents DataGrid is set up to use asynchronous replication. There is no option to use synchronous replication. Asynchronous replication allows you to run tests using a single cache agent. With asynchronous replication, the inference agent writes to a cache agent and returns. The cache provider then makes a separate call to another cache agent to make the replica. This means that the writes from the inference agent do not incur the cost of synchronous replication, because replication happens on a different thread in the background. However, a small window exists in which the inference agent has written to the cache, and the cache provider has not replicated the data yet. If the cache agent fails at this point, data is lost because there is no replica. To safeguard the data, use a backing store with cache-aside database write strategy. Oracle Coherence Uses Synchronous Replication With Oracle Coherence as the cache provider, TIBCO BusinessEvents is set up to use synchronous replication. There is no option to use asynchronous replication. With synchronous replication, when the inference agent writes to the cache, the cache provider makes a network call to another cache agent and makes a replica (or replicas) and then the call from the inference agent returns. This means making two cache puts for each cache operation. Therefore, synchronous replication is slower than asynchronous replication. You must also ensure that the required number of cache agents is always up and running (depending on the backup count).