A new paradigm for high availability and efficiency in replicated distributed databases

The paper presents a new paradigm for replication. Its major goal is to achieve performance similar to systems that do not employ replication and, at the same time, to offer the availability benefits that result from replication. The paradigm contributes two mechanisms. The first mechanism is an extended location service, for which it uses a logically centralized implementation. In addition, it modifies the traditional transaction-processing mechanism to interact with the location service inexpensively during transaction execution. The second mechanism is a priority-based, preemptive concurrency control algorithm which allows locks to be synchronously acquired at only a single replica. In addition, the paradigm exhibits desirable availability characteristics, satisfies the one-copy serializability correctness criterion and is easy to implement. For these reasons it is presented as a basis for designing efficient and highly available distributed databases