The notion of a timed register and its application to indulgent synchronization

• Authors: Michel Raynal;Gadi Taubenfeld
• Venue: Proceedings of the nineteenth annual ACM symposium on Parallel algorithms and architectures
• Year: 2007

A new type of shared object, called timed register, is proposed and used to design indulgent timing-based algorithms. A timed register generalizes the notion of an atomic register as follows: if a process invokes two consecutive operations on the same timed register which are a read followed by a write, then the write operation is executed only if it is invoked at most d time units after the read operation, where d is defined as part of the read operation. In this context, a timing-based algorithm is an algorithm whose correctness relies on the existence of a bound Δ such that any pair of consecutive constrained read and write operations issued by the same process on the same timed register are separated by at most Δ time units. An indulgent algorithm is an algorithm that always guarantees the safety properties, and ensures the liveness property as soon as the timing assumptions are satisfied. The usefulness of this new type of shared object is demonstrated by presenting simple and elegant indulgent timing-based algorithms that solve the mutual exclusion, l-exclusion, adaptive renaming,test&set, and consensus problems. Interestingly, timed registers are universal objects in systems with process crashes and transient timing failures (i.e., they allow building any concurrent object with a sequential specification). The paper also suggests connections with schedulers and contention managers.