Processing very “strong” event streams (with lots of events) to extract meaningful information in the shortest possible time (lowest latency) is the nontrivial task of complex event processing (CEP) systems. Part of the strategy to minimize processing time (and, thus, latency) is the clever identification of subexpressions in the queries that are evaluated only once with their result being reused in later parts of the overall query processing.

The authors present a novel “triaxial” hierarchy of how to identify more subexpressions than has been reported in previous algorithms by using (1) pattern matching, (2) concept hierarchies (a form of generalization and specialization), and (3) query operator hierarchies. A suitable “multi-query optimization strategy” then determines the best (that is, most cost efficient) strategy for actually executing and answering a particular query. All of this is presented in the required technical depth, including definitions and pseudocode algorithms to also enable others to implement this approach.

Additionally, a comprehensive set of benchmarks of the new approach against E-cube and other algorithms under different conditions (for example, query complexity, rates, or burstiness of the input event stream) shows varying but significant (from 22 percent to 100 percent) improvements. Other than using a medical, hospital-related domain for demonstrating the application of the method, the algorithm, of course, is wholly domain-independent.

The paper is quite technical and geared toward researchers in academia and business alike who are really interested in learning the programmatic inner workings of CEP algorithms.