Competitive Freshness Algorithms for Wait-free Data Objects

Rapport, 2005

Wait-free concurrent data objects are widely used in multiprocessor systems and real-time systems. Their popularity results from the fact that they avoid locking and that concurrent operations on such data objects are guaranteed to finish in a bounded number of steps regardless of the other operations interference. The data objects allow high access parallelism and guarantee correctness of the concurrent access with respect to its semantics. In such a highly-concurrent environment, where write-operations update the object state concurrently with read-operations, the age/freshness of the state returned by a read-operation is a significant measure of the object quality. In this paper, we first propose a freshness measure for wait-free concurrent data objects. Subsequently, we model the freshness problem as an online problem and present two algorithms for the problem. The first one is an optimal deterministic algorithm with freshness competitive ratio $\sqrt{\alpha}$, where $\alpha$ is a function of execution-time upper-bound of wait-free operations. The second one is a competitive randomized algorithm with freshness competitive ratio $\frac{\ln \alpha}{1 + \ln 2 - \frac{2}{\sqrt{\alpha}}}$.