There are three different strategies to employ when setting up replica servers.
With single-master replication, only one master server for any given partition or non-partitioned network holds writable copies of directory entries. Any replica servers have read-only copies of the directory entries. While both replicas and masters can perform searches, compares, and bind operations, only the master server can perform write operations.
The potential disadvantage to the single-master replication strategy is that the master server is a single point of failure. If the master server goes down, none of the replicas can process write operations.
The floating-master strategy is similar to the single-master strategy in that there is only one master server with write capabilities at any given time for a given partitioned or non-partitioned network. However, when implementing the floating-master strategy, when the master server goes down, a replica is automatically transformed into a master server by way of an algorithm.
One potential disadvantage to the floating-master replication strategy is that if your network becomes partitioned and replicas on either side of the partition become masters, the process of reconciling the new masters can be very complicated if the network is rejoined.
With multi-master replication, there are multiple master servers with their own read-write copies of the directory entry data. While the multi-master strategy eliminates the problem of having a single point of failure, update conflicts can occur between servers. In other words, if an entry's attribute is modified around the same time on two masters, an update conflict resolution policy, such as “last writer wins,” must be in place.
For information about how to set up replica servers, refer to the Administration Guide for the version of Sun Java System Directory Server that you are using.