Brief announcement: efficient implementation of a byzantine data storage system

We study the problem of implementing a replicated data store in an asynchronous system with an unbounded number of clients in which servers are subject to Byzantine failures and we seek to provide atomic access semantics for non selfverifying data. Implementations of shared registers with atomic semantics on servers subject to Byzantine failures have been proposed by a number of researchers. These solutions typically use Byzantine quorum systems and provide safe semantics. To provide stronger semantics, classical implementation of atomic registers can be used if the number of clients is bounded. Martin et al. [1] presented the first solution that provides atomic semantics for non self-verifying data in an asynchronous system with an unbounded number of readers and writers and in which servers are subject to Byzantine failures. Our solution is a significant improvement over previously proposed solutions: (1) In contrast to all previous solutions, the size of the timestamps in our solution is bounded by the number of operations and cannot be made arbitrarily large by Byzantine servers as is the case for all previous solutions; (2) It requires O(max|Q|), |Q| ≤ n, space to be used by readers compared to O(nf) space required in [1]; (3) In [1], each writer writes to all servers in the system which results in a write load equal to 1. In our solution, both read and write are to general Byzantine quorum sets which enable us to use quorums with low load and achieve a load for write