Meeting Agenda
Meeting Agenda
If you want more light at a terminal, please use one of the small desk lamps. If there aren't enough desk lamps, let us (cgl directors) know.
DEPARTMENT OF COMPUTER SCIENCE
UNIVERSITY OF WATERLOO
SEMINAR ACTIVITIES
Distributed Systems Seminar
- Friday, August 7, 1998
Paul A.S. Ward, Ph.D. Student, Department of Computer
Science, University of Waterloo, will speak on ``On the
Scalability of Monitoring and Debugging Distributed
Computations: Vector Clock Size''.
ROOM: Davis Centre Room DC1304
TIME: 2:30 - 3:30 p.m.
ABSTRACT:
A significant problem in monitoring and debugging
distributed computations is efficiently comparing the
precedence of events. To achieve this is constant time
a vector clock is associated with each event in the
computation. The cardinality of this vector clock is
equal to the number of concurrent entities in the
computation, which is not scalable. The theoretical
bound on the cardinality of the vector clock is the
dimension of the partial order that models the
computation. We therefore decided to investigate the
dimension of actual distributed computations. Since
this is an NP-hard problem, we developed approximate
algorithms to put a bound on the dimension. This talk
will describe the algorithms that were developed and
the results that have been found.
--------------------------------------------------------------------------
Distributed Systems Seminar
- Friday, August 7, 1998
Paul A.S. Ward, Ph.D. Student, Department of Computer
Science, University of Waterloo, will speak on ``On the
Scalability of Monitoring and Debugging Distributed
Computations: Vector Clock Size''.
ROOM: Davis Centre Room DC1304
TIME: 3:00 - 4:00 p.m. (NOTE CHANGE IN TIME)
--------------------------------------------------------------------------
DEPARTMENT OF COMPUTER SCIENCE
UNIVERSITY OF WATERLOO
SEMINAR ACTIVITIES
Master's Presentation
- Wednesday, August 5, 1998
Glenn Evans, Master's Student, Department of Computer
Science, University of Waterloo, will speak on
``Efficient Monte Carlo Global Spectral Illumination''.
ROOM: Davis Centre Room DC1304
TIME: 4:00 p.m.
ABSTRACT:
The importance of spectral rendering is increasing as
rendering algorithms attempt to heighten their
perceived level of realism. Wavelength dependent
effects can play a significant role in light
interactions with matter, through polarization,
wavelength dependent refraction, and body absorption
effects. However, the inclusion of spectral quantities
increases the complexity and can change the fundamental
design of a rendering algorithm.
Current research concentrates on using simplified
computational models to perform spectral rendering.
While computationally sound, most of simplified models
make assumptions that are physically incorrect in even
the simplest of situations. A more general and
efficient approach is needed to accomplish the spectral
rendering task without requiring complex data
structures and preprocessing algorithms.
Efficiency optimizations can be used to convert Monte
Carlo spectral point sampling into an efficient,
unbiased, and general technique for spectral global
illumination. The new algorithm eliminates many of the
restrictions of other spectral renders and increases
physical correctness.
The computational efficiency of the new algorithm is
shown to be an order of magnitude better than that of
Crude Monte Carlo and Quasi Monte Carlo approaches. A
hybrid algorithm is presented that improves the
correctness of previous Monte Carlo algorithms. Similar
efficiency optimizations can be used to efficiently
integrate spectral rendering into other Monte Carlo
algorithms.
August 4, 1998