An efficient and highly available read-one write-all protocol for replicated data management

A read-one write-all (ROWA) protocol for replicated data that allows a system to adjust to failures dynamically in order to keep the data available is proposed. If failures arrive mostly sequentially, the protocol keeps the data available as long as there is at least one operational replica. This is achieved by making the epoch mechanism, previously applicable to non-ROWA schemes only, usable within the ROWA discipline. Also, adjusting the system to a new configuration in the protocol is done completely asynchronously with reads and writes. In contrast, in the existing dynamic schemes (both ROWA and non-ROWA), system reconfiguration may interfere with and delay user transactions