Home
|
Course Outline
|
Lectures
|
Homework
|
Files
Class Room: 222 Natural Sciences Complex Class Hours: Mon Wed Fri 5:00 - 5:50 p.m. Home Page: http://www.physics.buffalo.edu/phy516/ Instructor: Dr. Richard J. Gonsalves Office Hours: Tue Thu 10:00 - 11:00 a.m. in 323 Fronczak; or by appointment. Telephone: (716) 645-2017 or 645-3043 Email: phygons@buffalo.edu
This course will cover selected Topics in High Performance Computing including cellular automata, finite element methods, molecular dynamics, Monte Carlo methods, and multigrid methods, with applications to classical fields, fluid dynamics, materials properties, nanostructures, and biomolecules. Sample applications will be developed in C/C++ and Fortran 90/95 extended with OpenMP directives and MPI functions. The choice of topics and methods will be guided by the following considerations which drive high performance computing:State-of-the-art Hardware and Tools
This course is sponsored by UB's Center for Computational Research which has HPC resources you can use. Supercomputing sites compete twice a year to make the Top 500 list. You should be familiar with HPC hardware, operating systems, and compilers from HPC I. Some useful references: Software Optimization for High Performance Computing, Wadleigh and Crawford (Prentice Hall 2000); Parallel Programming in OpenMP, R. Chandra et al. (Morgan Kaufmann 2000); Using MPI, W. Gropp et al. (MIT Press 1999).Tomorrow's Hardest Problems
For guidance you can look to yesterday's and today's hardest problems. The 1980's saw the development of massively parallel computers: the book Parallel Computing Works, which is also available online, describes many of the frontier problems solved on these machines. HPC frontiers in the 1990's were defined by Grand Challenge problems in science and engineering. Current high performance computing initiatives at federal agencies include SciDAC at the DoE, and ITR at NSF. The magazine Computing in Science and Engineering has accessible descriptions of recent HPC applications in a variety of fields.Top 10 Algorithms of the Century
New algorithms often allow dramatic performance gains in computing. What new algorithms will emerge during the twenty-first century? For guidance one might examine The Top 10 Algorithms of the twentieth century. A good reference on algorithms is "Numerical Recipes: The Art of Scientific Computing" in various languages by Press et al., also available online. There is much useful online material on algorithms, for example Templates for linear algebra solvers, and MGNet for multigrid applications. Numerous useful links can be found on the NHSE Software and HPCC Related Software pages.Visualization: The Human-Computer Interface
"The purpose of computing is insight, not numbers" - R.W. Hamming. The human visual system, among the most powerful information processing systems known, is often the final stage in HPC applications. Three-dimensional computer graphics can be very effective in the interpretation of large data sets generated by in high performance computing applications. The Visualization Toolkit will be used for some applications in this course.
High Performance Computing I
This course is the second semester of the two-semester High Performance Computing sequence. It assumes that you have taken HPC I, Fall 2001, or HPC I in a prior semester.Mathematical Background
You will need to be familiar with ordinary and partial differential equations at the level of MTH 418 or an equivalent course.Programming Experience
You should be proficient with programming in C or Fortran, and familiar with OpenMP and MPI.Please talk to me if you lack one or more of these prerequisites.
Homework
There will be several homework assignments during the semester. These may involve pencil and paper exercises on theory or algorithms, modifying and running sample code, writing code from scratch for serial and parallel execution, and a brief 2-3 page writeup of the results. The homework assignments will count for 60% of your final grade. Assignments will have definite due dates and late submissions will not be accepted.Semester Project
A semester project is required and will count for 40% of your grade. You may choose from a list of topics to be provided, or suggest a topic yourself. A one or two page topic proposal will be due Friday February 15, and your completed project report will be due Friday May 10. For the project you should
- research the necessary scientific or engineering background on the topic,
- develop code that will run on a serial computer using the most efficient algorithms available,
- adapt your code to one or more parallel computing environments and measure performance gains,
- and document what you have learned in the style of a journal publication in your field.
CCR Colloquium
Students registered for this course are expected to attend the CCR Colloquium talks. There will be approximately two talks scheduled each month by external and internal speakers. Lecture topics and homework assignments may reflect material presented in these colloquia.
If you have a disability, (physical or psychological) and require reasonable accomodations to enable you to participate in this course, such as note takers, readers, or extended time on exams and assignments, please contact the Office of Disability Services, 25 Capen Hall, 645-2608, and also see me during the first two weeks of class. ODS will provide you with information and review appropriate arrangements for reasonable accomodations.
Classes begin ... Tue Jan 22 Last day to add or drop courses ... Fri Feb 1 Last day to resign with "R" grade ... Fri Mar 22 Spring Recess ... Mon Mar 25 - Fri Mar 29 Last day of classes ... Fri May 3 Final examination period ... Sat May 4 - Thu May 9
Questions or comments:
phygons@buffalo.edu
Last updated:
Wednesday, 21 November 2001, 18:36:38