When data consistency is not a requirement, which SRDF mode is preferred when replicating large volumes of data?