MOHAMED A. SHARAF (Pitt/CS)

PhD Defense

Friday, June 22nd, 2007
12:00pm - SENSQ 6106

Abstract

Data streams processing is an emerging research area that is driven by the growing need for monitoring applications. A monitoring application continuously processes streams of data for interesting, significant, or anomalous events. Such applications include tracking the stock market, real-time detection of disease outbreaks, and environmental monitoring via sensor networks. Efficient employment of those monitoring applications requires advanced data processing techniques that can support the continuous processing of continuous rapid data streams. Such techniques go beyond the capabilities of the traditional store-then-query Data Base Management Systems. This need has led to a new data processing paradigm and created a new generation of data processing systems, supporting continuous queries (CQ) on data streams.

Primary emphasis in the development of first generation Data Stream Management Systems (DSMSs) was given to basic functionality.
However, in order to support large-scale heterogeneous applications that are envisioned for subsequent generations of DSMSs, greater attention will have to be paid to performance issues. Towards this, the proposed thesis will introduce new algorithms and metrics to the current design of DSMSs.

This thesis identifies a collection of quality of service (QoS) and quality of data (QoD) metrics that are suitable for a wide range of monitoring applications. The establishment of well-defined metrics aids in the development of novel algorithms that are optimal with respect to a particular metric. Our proposed algorithms exploit the valuable chances for optimization that arise in the presence of multiple application. Additionally, they aim to balance the trade-off between the DSMS's overall performance and the performance perceived by individual applications. Furthermore, we provide efficient implementations of the proposed algorithms and we also extend them to exploit sharing in optimized multi-query plans and multi-stream CQs. Finally, we experimentally show that our algorithms consistently outperform currently used ones.

Dissertation co-Advisers

Prof. Panos K. Chrysanthis, Department of Computer Science
Prof. Alexandros Labrinidis, Department of Computer Science

Committee Members

Prof. Walid Aref, University of Purdue
Prof. Christos Faloutsos, Carnegie Mellon University
Prof. Kirk Pruhs, Department of Computer Science