Uniform TitleScalable and robust stream processing
NameShkapenyuk, Vladislav (author), Muthukrishnan, Shanmugavelayutham (chair), Iftode, Liviu (internal member), Marian, AmÃ©lie (internal member), Srivastava, Divesh (outside member), Rutgers University, Graduate School-New Brunswick,
Data transmission systems
DescriptionDistributed Data Stream Management Systems (DSMS) are increasingly used for the processing
of high-rate data streams in real-time. An effective query optimization mechanism is a critical component that allows DSMS to deal with extreme data rates and large numbers of long-running concurrent queries. This dissertation investigates how to utilize semantic query analysis to perform query optimizations that enable scalable and robust data stream processing. We address three technical challenges faced by streaming system: (1) monitoring and correlating large number of diverse data streams with significant variations in data rates; (2) the ability to remain stable and produce correct answers even under overload conditions, and (3) supporting efficient distributed query processing to easily scale with increases in the number of processing nodes and stream data rates.
First, we propose a heartbeat mechanism to prevent the DSMS from blocking when some of the monitored streams temporarily stall or slow down. By generating special punctuation messages at low-level query nodes and propagating them throughout the entire query execution plan, our heartbeat mechanism effectively unblocks all stalled query nodes.
The second contribution of this dissertation addresses the problem of DSMS robustness when a load on a system increases by orders of magnitude. We introduce a query-aware sampling mechanism for guaranteeing the system's stability and the correctness of its query output under overload conditions. The mechanism is generic and supports arbitrary complex query sets.
Finally, we address the problem of scalable distributed evaluation of streaming queries. The key contribution of the dissertation is a query-aware partitioning mechanism that allows us to scale the performance of the streaming queries in a close to linear fashion. We propose a query analysis framework for determining the optimal partitioning and a partition-aware distributed query optimizer that takes advantage of existing partitions.
In summary, the contributions made by this dissertation in the area of streaming query
optimization enable Data Stream Management Systems to scale to extreme data rates, gracefully handle overload conditions and support a large number of diverse input streams, enabling industrial-scale applications of DSMS technology.
Note[bibliography] Includes bibliographical references (p. 156-165).
CollectionGraduate School - New Brunswick Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.