The DYNAQUEST-Framework for Dynamic and Adaptive Source Selection

摘要

Information overload is one of the main reasons why efficient use of internet information sources is a mostly cumbersome process. Closely connected to this general observations is the problem of resource discovery: Which are the best information sources to answer my query respecting my preferences in the current execution context (e.g. time of day). Virtual databases try to free users from the burden of distribution and heterogeneity regarding the data access and data quality of information source in the Internet. Queries are formulated against one (or more) virtual views, relevant sources are determined by the system, queried, and finally the results of these sources are combined. A latent problem in open systems like the Internet is caused by their high dynamics: A previously fast answering server can be unreachable or new and better-suited sources may be available. To cope with such problems, intelligent and fault-tolerant mechanisms are needed. DYNAQUEST, a framework for dynamic and adaptive query processing, is our proposal to build robust virtual database systems. This agent-based approach helps locating relevant sources based on user preferences concerning data and quality aspects. While processing a query with these selected sources, errors (e.g. a source failure) are caught and DYNAQUEST tries to recover the query instead of breaking it (e.g. by choosing an alternate source). Our approach always tries to comply with desired information quality demands while optimizing or adapting query processing.