Current mediator-wrapper architectures do not have the flexibility to scale to multiple Web wrapped sources, where some sources may be redundant, and some sources may provide incomplete answers to a query. Also, since current mediators do not export an interface reflecting the capabilities and contents of the sources, they are not able to perform CBR with respect to other mediators. This can be critical when the number of available sources dramatically increases and the CBR task has to be performed for more than one mediator. We propose a mediator which operates within a particular domain and the sources/wrappers that have registered within this domain. It will consider the query capability and the contents (domains) of these sources, and will provide a single interface to the other mediators. It reflects the capabilities and contents of the sources/wrappers. Source descriptions specify the content and query capability of the sources. These are used to derive the single interface and to decide which queries can be accepted. Sources are partitioned into equivalence classes, based on their descriptions. These equivalence classes are partially ordered, and the lattices that correspond to these orderings are used to identify the relevant sources for an accepted query. If there is redundancy of the sources, alternate sources are identified for queries.

Back to the Summer 1998 dbchat index