Thursday October 10, 2002
Noon - 1:00 p.m.
Hamerschlag Hall, D210
Assistant Professor, UC Berkeley
Griffin: Towards an
Agile, Predictive Infrastructure
In this talk, I will present the Griffin project, the goals, and recent
results and achievements. We are exploring the development of a network
analysis and emulation architecture that provides developers with a highly
accurate emulation of a wired or wireless link. Griffin also provides
applications with predictive information about network conditions and
topologies that enables them to adapt BEFORE conditions deteriorate or
improve. Preliminary results show that Griffin techniques can help applications:
reduce wide-area bandwidth requirements by up to 75%, improve performance
on slow networks by up to 90%, and significantly improve performance over
The Griffin architecture includes components at several levels of the
network stack, from link and media access layers, through overlay networks,
and up to applications. The first half of the talk provides an overview
of Griffin, our design goals, architecture, and high-level components.
The second half of the talk will focus on the detailed design of one of
the components, Tapas, Griffin's network monitoring, modeling, emulation,
and prediction system. I will present results from a detailed analysis
of several existing and new networks, including IP, GSM, and IEEE 802.11b
networks. I will also show how the Tapas techniques provide two significant
benefits to protocol and application developers by enabling the development
of new protocols using highly accurate simulation techniques, and simplifying
the development of predictive, adaptive applications.
Anthony Joseph received his Ph.D. in Computer Science with a Minor in
Management of Research and Technology from the Massachusetts Institute
of Technology in 1998. He is now an Assistant Professor at UC Berkeley
and has reseach interests in access to information.
Further Seminar Info: