Efficient processing of multiple nested event pattern queries over multi-dimensional event streams based on a triaxial hierarchical model

Objective: For efficient and sophisticated analysis of complex event patterns that appear in streams of big data from health care information systems and support for decision-making, a triaxial hierarchical model is proposed in this paper.

Methods and material: Our triaxial hierarchical model is developed by focusing on hierarchies among nested event pattern queries with an event concept hierarchy, thereby allowing us to identify the relationships among the expressions and sub-expressions of the queries extensively. We devise a cost-based heuristic by means of the triaxial hierarchical model to find an optimised query execution plan in terms of the costs of both the operators and the communications between them. According to the triaxial hierarchical model, we can also calculate how to reuse the results of the common sub-expressions in multiple queries. By integrating the optimised query execution plan with the reuse schemes, a multi-query optimisation strategy is developed to accomplish efficient processing of multiple nested event pattern queries.

Results: We present empirical studies in which the performance of multi-query optimisation strategy was examined under various stream input rates and workloads. Specifically, the workloads of pattern queries can be used for supporting monitoring patients' conditions. On the other hand, experiments with varying input rates of streams can correspond to changes of the numbers of patients that a system should manage, whereas burst input rates can correspond to changes of rushes of patients to be taken care of. The experimental results have shown that, in Workload 1, our proposal can improve about 4 and 2 times throughput comparing with the relative works, respectively; in Workload 2, our proposal can improve about 3 and 2 times throughput comparing with the relative works, respectively; in Workload 3, our proposal can improve about 6 times throughput comparing with the relative work.

Conclusion: The experimental results demonstrated that our proposal was able to process complex queries efficiently which can support health information systems and further decision-making.