CFP: Workshop on Performance Modeling - ICCS Melbourne 6/2/03
Dear Colleague,
You are invited to submit a workshop paper in the general topic area of
HPC performance benchmarking, modeling, and extrapolation. The workshop
will take place in Melbourne Australia June 2nd (see
http://www.science.uva.nl/events/ICCS2003/ for general conference information).
Preference will be given to papers with innovative and creative ideas
for dealing with the challenges inherent in evaluating current HPC
systems, and/or designing future ones guided by desires of the users
and computational demands of scientific applications. The workshop
emphasis is towards creativity and thinking "out-of-the-box". We
encourage submissions with partial research results and promising
lines of investigation even if these have not yet been thoroughly
developed.
SUBMISSION INSTRUCTIONS AND DEADLINES:
Please submit draft papers for consideration by January 15th, 2003 in
either .ps, .pdf, or MS Word format via email to
allans@sdsc.edu
Notification of acceptance February 1st.
Final versions due February 15th. Papers longer than 10 pages are not
encouraged but not specifically prohibited (you may be asked to trim
the final version).
TITLE:
Benchmarking, Performance Evaluation, and Performance Modeling
CHAIRS:
Allan Snavely, SDSC
Jeff Vetter, LLNL
DESCRIPTION:
This workshop will explore performance evaluation of HPC systems and
Grids with an emphasis on high order performance objective functions.
For example, while common wisdom has held that "you can't benchmark
user satisfaction" several interesting recent works exhibit the
contrary. It may be possible to design, build, and schedule systems in
ways that explicitly optimize for user's applications and the values
they attach to timely results. And it may be possible to do this in a
fully quantitative and objective way.
Specific areas of interest include (but are not limited to)
Correlating low-level benchmarks to real applications performance
Methods and uses for "application signatures" (abstract
characterizations of the computational demands of HPC applications)
Performance Modeling including:
- Designing systems to best support particular algorithm/application
families
- Innovative approaches to extrapolating the expected performance of
future machines
- Methods for picking the right machine for an application
Full systems benchmarking including:
- Benchmarking scheduler software and other HPC and Grid middleware
- I/O subsystem benchmarking
- Methods for "benchmarking" the users to ascertain their utility
functions
CHAIR CONTACT INFO:
Allan Snavely, Ph.D.
PMaC Laboratory Leader
Integrative Computational Sciences
SDSC University of California, San Diego
10100 Hopkins Drive, La Jolla
E-mail: allans@sdsc.edu
Phone: (858) 534-5158
Fax: (858) 534-5117
http://www.sdsc.edu/~allans
http://www.sdsc.edu/PMaC
Jeffrey Vetter, Ph.D.
Center for Applied Scientific Computing
Lawrence Livermore National Laboratory
Box 808, L-560
Livermore, CA 94551
vetter3@llnl.gov
Phone: (925) 424-6284
Fax: (925) 422-6287