In replicated mode, each piece of data (key) is replicated to every node of the cluster. As the key is available in all nodes, this approach offers really high availability. 

The following diagram shows the data replication. Here, each JVM is holding either a primary copy or a backup copy of each key. JVM 1 has the primary copy of Key 1 and a backup copy of Key 2 and Key 3. So, now, if we add another key (Key 4) to the cache, it will add the Key 4 primary copy to any of these three JVMs and keep two backups in the other two JVMs:

This approach offers the highest availability and read performance, but when you modify data or add a new key, it has to propagate the change to every node of the cluster. The write performance will slow down. This mode is not scalable, as adding a new node will not allow you to store more data.

Apache Ignite recommends that you use the replicated mode when the cache footprint is small and the data is updated infrequently.