At the DB level, it facilitates master-master replication set-up, eliminating the auto_increment collision problems. Master-master replication, in turn, allows building distributed applications that can handle net splits reasonably well.