Subsuming multiple sliding windows for shared stream computation

Shared evaluation of multiple user requests is an utmost priority for stream processing engines in order to achieve high throughput and provide timely results. Given that most continuous queries specify windowing constraints, we suggest a multi-level scheme for concurrent evaluation of time-based sliding windows seeking for potential subsumptions among them. As requests may be registered or suspended dynamically, we develop a technique for choosing the most suitable embedding of a given window into a group composed of multi-grained time frames already employed for other queries. Intuitively, the proposed methodology "clusters" windowed operators into common hierarchical constructs, thus drastically reducing the need for their separate evaluation. Our empirical study confirms that such a scheme achieves dramatic memory savings with almost negligible maintenance cost.