Type-specific replication algorithms for multiprocessors

The use of replication for enhancement of the availability of data in multiprocessor systems is discussed. The author explores theoretical aspects of the extent to which it is possible to exploit the semantics of data to make replication more effective. Some of the literature is surveyed, and some opinions about appropriate models and interesting open questions are presented. In particular, linearizability as an appropriate correctness criterion for concurrent objects, wait-free synchronization, and synchronous and asynchronous message-passing models are discussed.<
>