Consistency in a Stream Warehouse

A stream warehouse is a Data Stream Management System (DSMS) that stores a very long history, e.g. years or decades; or equivalently a data warehouse that is continuously loaded. A stream warehouse enables queries that seamlessly range from realtime alerting and diagnostics to long-term data mining. However, continuously loading data from uncontrolled sources into a realtime stream warehouse introduces a new consistency problem: users want results in as timely a fashion as possible, but “stable” results often require lengthy synchronization delays. In this paper we develop a theory of consistency for stream warehouses that allows for multiple consistency levels, we show how to restrict query answers to a given consistency level, and we show how warehouse maintenance can be optimized using knowledge of the consistency levels required by materialized views.