Storage sites are allocated pairs of nodes in which one node of the pair acts as an owner node while the other a backup node. When a local owner node receives a colliding write, the local owner node obtains a lock on the modified database and transfers both the write data and metadata to a remote owner node. The remote owner node returns a write complete message and the local site owner unlocks the modified database. When a local backup node receives a colliding write, the local backup node requests a lock from the local owner node and sends the write data to the remote owner node while the local owner node sends the write metadata to the remote owner node. The remote owner node then returns a write complete message to the local backup node which then requests the modified database be unlocked by the local owner node.