Hierarchical Detection of Strong Unstable Conjunctive Predicates in Large-Scale Systems

摘要

In large-scale systems where an on-going monitoring program is needed, the traditional predicate detection algorithms become undesirable due to their high overhead and inability to do repeated detection and to resume the detection after a node failure. This paper presents an on-line decentralized algorithm that detects strong conjunctive predicates in a large-scale system. Our algorithm assumes a preconstructed spanning tree in the system, and detects all satisfactions of the predicate in a hierarchical manner. When a node fails or moves and the structure of the spanning tree is changed, our algorithm is able to recover from this situation and continue the detection of further occurrences of the predicate satisfactions. The hierarchical structure of our algorithm also provides a finer-grained monitoring in those large-scale systems where grouping is established and the monitoring happens at the group level. Comparing with other detection algorithms, our algorithm incurs a low space and time cost, which is distributed across all the nodes in the system, and a low message complexity. Our algorithm is particularly beneficial to resource-constrained systems.