Multiversion divergence control of time fuzziness
Calton Pu, Miu K. Tsang, Kun‐Lung Wu, Philip S. Yu · 1994
Epsilon Serializability (ESR) has been proposed to manage and control inconsistency in extending the classic transaction processing. ESR increases system concurrency by tolerating a bounded amount of inconsistency. In this paper, we present multiversion divergence control (mvDC) algorithms that support ESR with not only value but also time fuzziness in multiversion databases. Unlike value fuzziness, accumulating time fuzziness is semantically different. A simple summation of the length of two time intervals may either underestimate the total time fuzziness, resulting in incorrect execution, or overestimate the total time fuzziness, unnecessarily degrading the effectiveness of mvESR. We present a new operation, called TimeUnion, to accurately accumulate the total time fuzziness. Because of the accurate control of time and value fuzziness by the mvDC algorithm, mvESR is very suitable for the use of multiversion databases for real-time applications that may tolerate a limited degree of data inconsistency but prefer more data recency.