Date: Wednesday, February 14, 2001
Time: 10:30 A.M. (Talk) <<<<<<<<
NOTE <<<<<<<<
Place: Withers 402-A, NCSU Historical Campus (click for
courtesy parking request)
Speaker: Beth Plale, BOPS Inc. and Computer Science, Duke U.
Dynamic Querying of Data Streams with the dQUOB System
Abstract: This talk is focused on managing the flow of data between communicating components in high performance parallel and distributed computations composed of interacting programs, actuators, sensors, remote data sources, distributed users, and high-end computational engines.
In response to the proliferation of Internet access and the continual drop in per-flop cost of cluster computers, high performance computations have scaled in multiple directions. There is a scaling of potential data sources: where heretofore access to a remote data source during computation was prohibitively slow and often with unwieldy interface issues, those barriers are now dropping. There is scaling in problem size: scientific models that previously taxed the resources of a supercomputer individually are now coupled. Finally, there is scaling in number and types of users.
These software programs have a common characteristic; that is, they exhibit significant data transfer amongst peer components. But a natural outcome of the recent scaling is a problem in mismatch between data supplier and data consumer, in terms of the amount of data, flow rate, and data representation. Further, it has raised the need for comprehensive solutions to satisfying an individual's specific information needs.
My research has been to develop a model of the event data transfer, and particularly a system, called dQUOB, by which one can create computational entities that manipulate data streams. I address the problem of data mismatch at the application level. That is, I assume the underlying wire format resolves issues like byte ordering and word size inconsistencies that exist amongst heterogeneous platforms, and focus instead on broader application-level issues such as "user X needs a subset of fields from the union of events A, B, and C," or "user Y needs units in PPM where the supplier publishes them in PPB". This talk is a discussion of the dQUOB approach to managing data streams.
Short Bio: Beth Plale is a Postdoc in the College of Computing at Georgia Institute of Technology. She received a PhD in Computer Science from State University of New York at Binghamton in 1998. Dr. Plale's research interests include managing information in large-scale data streams, parallel and distributed computing, run-time detection, database query processing, on-line monitoring and steering, and software safety.
Hosts: C.
Healey, Computer Science, NCSU