Data warehouses are becoming a powerful tool to analyze enterprise data. A critical demand imposed by the users of data warehouses is that the time to get an answer (latency) after posing a query is to be as short as possible. It is arguable that a quick, albeit approximate, answer that can be refined over time is much better than a perfect answer for which a user has to wait a long time. In this paper we addressed the issue of online support for data warehouse queries, meaning the ability to reduce the latency of the answer at the expense of having an approximate answer that can be refined as the user is looking at it. Previous work has address the online support by using sampling techniques. We argue that a better way is to preclassify the cells of the data cube into error bins and bring the target data for a query in “waves,” i.e., by fetching the data in those bins one after the other. The cells are classified into bins by means of the usage of a data model (e.g., linear regression, log-linear models) that allows the system to obtain an approximate value for each of the data cube cells. The difference between the estimated value and the true value is the estimation error, and its magnitude determines to which bin the cell belongs. The estimated value given by the model serves to give a very quick, yet approximate answer, that will be refined online by bringing cells from the error bins. Experiments show that this technique is a good way to support online aggregation.
Weitere Kapitel dieses Buchs durch Wischen aufrufen
- Supporting Online Queries in ROLAP
- Springer Berlin Heidelberg
Neuer Inhalt/© ITandMEDIA