Loading Events

Systems Seminar - CSE

XML Selectivity Estimation and Dynamic Data Caching

Sriram Padmanabhan

This talk will discuss our recent research activity in XML databases and dynamic data caching. I will begin with an overview of IBM's XML information integration architecture highlighting the research challenges. I will then present our work on incremental and adaptive selectivity estimation for XML query predicates. We use the Markovian property for estimating selectivities of longer paths from shorter paths. The selectivity estimates for the shorter paths are maintained and updated incrementally. We have studied a couple of update techniques for incremental maintenance and report on the accuracy of these techniques for workoads based on well-known XML data. This work was published in VLDB 2002.

The second part of the talk will describe DBProxy, which is a research prototype of a self-managing dynamic data cache primarily intended for web applications. Although caching and related topics are well studied in the database research literature, practical implementations of caching have tended to be very simple or requiring significant administration. Our approach to this problem is to use the semantic caching concepts but also enable data sharing, replacement, and reduced administration costs. I will highlight some of these features and present experimental results based on the TPC-W benchmark. This work will be published in ICDE 2003.

Sriram Padmanabhan is a Research Staff Member and Manager of the Scalable Data Management Group at the IBM TJ Watson Research Center, Hawthorne, NY. He joined the Watson Research Center in 1992 after completing his Ph.D from the EECS Department, The University of Michigan. His research interests include most aspects of database processing such as query processing, parallel databases, data warehousing, federated database processing, caching, and semistructured databases. In IBM, he was a core member of the Research team that prototyped and subsequently developed IBM's Shared-Nothing Parallel Database product, DB2 Parallel Edition. He continues to be engaged with new enhancements to the IBM DB2 product. Dr. Padmanabhan has published in leading database conferences and journals.

Sponsored by