[My email will be forwarded.]
|
[I'm not working at LBL any longer.] [My email will be forwarded.] |
|
April, 2002 to April, 2006, I worked at Lawrence Berkeley National Lab (LBL) in the Computational Research Division(CRD). I work in the Applied Numerical Algorithms group (ANAG) under Phil Colella, part of the CRD High Performance Computing Research Dept.
My work in ANAG focuses on algorithms and software for solving partial differential equations using solution-adaptive Cartesian grid methods, and applications thereof. Algorithm issues include particle-in-cell methods for adaptive grids, high-order accuracy, complex geometries and moving boundaries. Software issues include object-oriented design, portability and performance. Current applications include particle accelerator design, fusion modeling, cellular biology, early universe simlations, fluid dynamics. This work is part of the Applied Partial Differential Equations Center of the DOE SciDAC program.
ANAG has released a package of software called Chombo that is the result of much of the group's research. It can be found on the ANAG web page, along with other software produced by the group.
Previously, I worked in the NERSC Future Technologies Group (FTG), focusing on numerical algorithms for, and performance of, high-performance parallel computers, mainly clusters constructed from commodity systems, and in particular how they compare to traditional, vendor-constructed parallel supercomputers. I think that issues such as load balancing, job scheduling, availability and quality of compilers and libraries, network and I/O performance, etc are just as important as floating point arithmetic performance. The performance bottlenecks in most of the microprocessor-based systems I looked at were in the memory, disk and network subsystems, not in the CPU. In particular, the fastest Intel and Alpha-based systems at the time were faster than a Cray J90 on small, compute-intensive problems. But as soon as you try to run a real application for a realistic problem size, the advantages of the Cray start to show up. As I've said a lot: "Bandwidth costs money, flops are free". This is even more true in 2004 than it was in 1999. My work in FTG tried to analyze and address some of these issues.
In between working in FTG and my current position, I left the lab and worked in private industry for two years in computational finance, specifically developing real-time systems for computerized trading. This was an interesting experience with little in common with research.
My other interests range over most aspects of numerical computing, from low level issues like system architecture and programming language design to high level issues like numerical algorithm design for differential equations and nonlinear optimization, and many things in between.
Before NERSC, I was a graduate student at Rice University in the Computational and Applied Math department where I received a Ph.D. in Computational Science and Engineering. My thesis advisor was Prof. John Dennis. My thesis topic was nonlinear optimization using approximations (aka models, aka surrogates) (see below). The target applications were functions that are so expensive to compute that normal optimization methods are impractical. An interesting recent application of this is described in "Optimal Aeroacoustic Shape Design Using the Surrogate Management Framework" by Marsden, et al.
I also worked on the Parallel Direct Search (PDS) nonlinear optimization code, a portable parallel program for optimization without computing derivatives. The great thing about PDS is it will work on almost any function and it can use as much compute power as you can give it. PDS can be tested on the net using the Network Enabled Optimization System (NEOS) server at Argonne National Laboratory. [NOTE: PDS/NEOS isn't working currently.] The source code is available. Send me email to get it. PDS has been superceded to a large degree by the APPSPACK and NOMAD packages.
In a previous life, I worked as a consultant in CFD and high performance computing for Rose Engineering and Research and for various customers at NASA (Ames, Lewis and Langley). Mostly I worked on simulations of high speed (Mach 4--18) engine inlet systems (e.g. this Mach 5 inlet in the 10'x10' supersonic tunnel at NASA/Lewis). I also simulated the insides of a couple of the wind tunnels at NASA/Ames ( Unitary 9'x7' and 11'). and the outside of a Boeing 747SP (for the SOFIA project). I wrote code for everything from PDE solutions to data analysis to graphics to networked printers.
I also worked for Thinking Machines Corp. during the CM-2 days. This is where I learned to love Lisp and hack on Symbolics Lisp Machines (the best software development environment ever created).
Long ago I went to school at MIT where I got a B.S. in Aeronautical Engineering. I spent most of my time working for various theater groups around campus doing sets and lighting (and occasionally did some class work). I wrote my senior thesis on a wind tunnel study of an ultralight aircraft wing.
A Framework for Managing Models in Nonlinear Optimization of Computationally Expensive Functions
Abstract: (HTML),(Postscript, 185KB, 2pgs)
Whole document: (Postscript, 1.2MB, 185pgs),
(gzip'd PS, 310KB)
A.J. Booker, J.E. Dennis, Jr., Paul D. Frank, David B. Serafini,
Virginia Torczon, Michael W. Trosset,
A Rigorous Framework for Optimization of Expensive Function by Surrogates
(PostScript, 400KB, 24pgs),
(gzip'd PS, 94K)
Structural Optimization, v17, n1, pp1--13.
A.J. Booker, J.E. Dennis, Jr., Paul D. Frank, Douglas W. Moore, David B. Serafini,
Managing Surrogate Objectives to Optimize a Helicopter Rotor Design - Further experiments
(PostScript, 113KB, 7pgs),
(gzip'd PS, 36K)
(AIAA paper 98-4717, presented at the 7th AIAA/USAF/NASA/ISSMO Symposium on
Multidisciplinary Analysis & Optimization, St.Louis, MO, Sept. 1998.)
A.J. Booker, J.E. Dennis, Jr., P.D. Frank, D.B. Serafini, V. Torczon
Optimization Using Surrogate Objectives on a Helicopter Test
Example
in Optimal Design edited by J. Burns and E. Cliff,
SIAM, Philadelphia 1998.
SciDAC project, APDEC review meeting, Dec 3, 2003
Title: Review of AMR/PIC project.
Particle-in-cell methods in the Chombo Adaptive Mesh Refinement framework.
Dept. of Energy Office of Industrial Technology Multiphase Fluid Dynamics
Research Consortium, 1st annual meeting (July 1998)
Title: Parallelization and Visualization of the MFIX flow code
gzip'd Postscript (7 pages, 70KB)
2nd Annual Conference on Minorities & Applied Mathematics (Sept 1998)
Title: A View from the Front Lines: Computational Mathematics in the "Real" World
gzip'd Postscript (21 pages, 200KB)
gzip'd Postscript, 4 slides per page (6 pages, 200KB)
SGI Showcase file (70KB)
I'm more than happy to receive mail about any of the above topics.
dbs (at) hpcrd (dot) lbl (dot) gov