Extending data warehouses by semiconsistent views

The extremely expensive process of integrating data to achieve a consolidated database is one of the main problems in building a data warehouse. Since operational data sources may exhibit a high update rate, the data warehouse information is out of date by the order of magnitude. To avoid out-dated information the alternative is to query the data sources directly, which results in higher query runtimes or in the complete failure producing a result if data sources are currently not available. This paper discusses an approach to close this gap: multiple snapshots of participating data sources are cached in a middleware layer and user queries are routed to this set of snapshots which are providing an almost consistent global database view. We describe the architecture of our information middleware approach, develop different join semantics to combine different data sources, and propose algorithms for picking time consistent cuts in the history of local snapshots.