Patent attributes
Methods and apparatuses are described to provide a mechanism to improve replication performance by understanding available bandwidths of multiple source servers and load balancing the replication among the source servers based on fingerprints of data units to be replicated. A separate source server can be identified for one source server to replicate the data units to a target server. A replication request is sent from the one source server to the separate source server to identify the data units. A reply may be received from the separate source server to indicate that one or more of the data units are replicated from the separate source server. Subsequently, the remaining ones of the data units not indicated in the reply may be sent directly from the one source server to the target server without sending the one or more data units indicated in the reply.