Adaptive Query Processing

As the area of data management for the Internet has gained in popularity, recent work has focused on effectively dealing with unpredictable, dynamic data volumes and transfer rates using adaptive query processing techniques.  Important requirements of the Internet domain include: (1) the ability to process XML data as it streams in from the network, in addition to working on locally stored data; (2) dynamic scheduling of operators to adjust to I/O delays and flow rates; (3) sharing and re-use of data across multiple queries, where possible; (4) the ability to output results and later update them.  An equally important consideration is the high degree of variability in performance needs for different query processing domains: perhaps an ad-hoc query application should optimize for display of incomplete and partial incremental results, whereas a corporate data integration application may need the best time-to-completion and may have very strict data "freshness" guarantees.  The goal of the Tukwila project is to design a query processing system that supports a range of adaptive techniques that are configurable for different query contexts.  Recent work focuses on providing "previews" to query answers -- these previews may be approximate answers, subsets of the real answers, or other results that provide early feedback to the user.

Project Members

Related Projects