Capabilities-Based Query Rewriting in Mediator Systems

Users today are struggling to integrate a broad range of information sources providing different levels of query capabilities. Currently, data sources with different and limited capabilities are accessed either by writing rich functional wrappers for the more primitive sources, or by dealing with all sources at a "lowest common denominator". This paper explores a third approach, in which a mediator ensures that sources receive queries they can handle, while still taking advantage of all of the query power of the source. We propose an architecture that enables this, and identify a key component of that architecture, the Capabilities-Based Rewriter (CBR). The CBR takes as input a description of the capabilities of a data source, and a query targeted for that data source. From these, the CBR determines component queries to be sent to the sources, commensurate with their abilities, and computes a plan for combining their results using joins, unions, selections, and projections. We provide a language to describe the query capability of data sources and a plan generation algorithm. Our description language and plan generation algorithm are schema independent and handle SPJ queries. We also extend CBR with a cost-based optimizer. The net effect is that we prune without losing completeness. Finally we compare the implementation of a CBR for the Garlic project with the algorithms proposed in this paper.