From: Danny Wyatt (danny@cs.washington.edu)
Date: Mon May 03 2004 - 09:52:15 PDT
I find it comforting that algorithms do exist for synthesizing queries
from views, provided they are all expressed as conjunctive queries
without negation or comparison. That leads to the obvious next step of
integrating disparate sources by describing each source using logical
views and automatically merging these views to answer other queries. Of
course, that assumes that all of the views are logically consistent and
that each source admits a description as a conjunctive query.
The next challenge, then, seems to be divining such descriptions and
managing inconsistency and contradiction between sources. IM requires
that all integrated sources be described in the terms of one schema,
grounding them all in universal set of predicates and thereby allowing a
single mediator to handle all queries. Tsimmis allows multiple
mediators to each specify their own schema which makes integrating new
sources easier but sacrifices uniformity of queries unless a global
mediator is imposed on top. I think this points out a fundamental
trade-off in information integration: the balance between local and
global semantics and the need for extensible integration while still
supporting querying over the entire multisource DB (or KB).
Overall, like much of Ullman's work, I liked this paper, though I agree
it could have made its organization more plain.
This archive was generated by hypermail 2.1.6 : Mon May 03 2004 - 09:52:21 PDT