University of Toronto
Department of Computer Science

A Distinguished Lecture on Computer Science

Joseph M. Hellerstein

University of California, Berkeley

Adaptive Dataflow: Eddies and Rivers

Abstract

Data movement is a central theme in modern computer systems. One tradition in data movement comes from the networking community, which has been responsible for the low-level delivery of uninterpreted messages across unpredictable wide-area networks. Another tradition comes from the database community, which developed architectures to flow data through high-level operators on "shared-nothing" clusters of computers, and to minimize data movement across the wide area. Traditionally, the networking solutions have been very adaptive to changes in performance of the network, but have paid little attention to the contents of the data or the software using it. By contrast, the database solutions have taken great advantage of understanding both the data and the operators through which it flows, but have been relatively rigid in adapting to volatility in performance.

At Berkeley, we have been working on new architectures for adaptive dataflow, which intelligently route data through high-level operators on a network, and adapt to performance volatility at a very fine grain. In this talk, I present Eddies and Rivers, two adaptive dataflow mechanisms we are using in the Telegraph system at Berkeley. Eddies adapt to volatility by flexibly routing data through operators in a dataflow; Rivers adapt by flexibly balancing data loads across machines in a network. I will discuss how these ideas are being used in the Telegraph system we are developing, for scenarios including sensor networks and a novel web query system. I will also raise a number of research challenges that we face in achieving fully adaptive dataflow systems.

Biography

Joseph M. Hellerstein is an associate professor of computer science at the University of California, Berkeley. He is an Alfred P. Sloan Research University of California, Berkeley. He is an Alfred P. Sloan Research Fellow, a recipient of the NSF CAREER award, and was recently named one of the top 100 young technology innovators by MIT's Technology Review magazine. Prof. Hellerstein's research centers on database and information management systems, with a focus on federated and adaptive information systems, interactive analysis and transformation of massive datasets, and generalized indexing and indexability. Hellerstein is the Chief Scientist and co-founder of Cohera Corporation, a vendor of electronic catalog software based on federated data management technology. He received his Ph.D. from the University of Wisconsin, Madison, a master's degree from UC Berkeley, and a bachelor's degree from Harvard.


Host: contact Prof. Miller for information on the speaker's schedule.

Time and Location: return to the 2000 Colloquia Series main page.


R. J. Miller
Last modified: Mon Aug 28 20:11:03 EDT 2000