Atomic Register Abstractions for Byzantine-Prone Distributed Systems

The construction of the atomic register abstraction over crash-prone asynchronous message-passing systems has been extensively studied since the founding work of Attiya, Bar-Noy, and Dolev. It has been shown that t < n/2 (where t is the maximal number of processes that may be faulty) is a necessary and sufficient requirement to build an atomic register. However, little attention has been paid to systems where faulty processes may exhibit a Byzantine behavior. This paper studies three definitions of linearizable single-writer multi-reader registers encountered in the state of the art: Read/Write registers whose read perations return the last written value, Read/Write-Increment registers whose read perations return both the last written value and the number of previously written values, and Read/Append registers whose read perations return the sequence of all previously written values. More specifically, it compares their computing power and the necessary and sufficient conditions on the maximum ratio t/n which makes it possible to build reductions from one register to another. Namely, we prove that t < n/3 is necessary and sufficient to implement a Read/Write-Increment register from Read/Write registers whereas this bound is only t < n/2 for a reduction from a Read/Append register to Read/Write-Increment registers. Reduction algorithms meeting these bounds are also provided.