CSE 262 Class Projects (Spring 2011)

One of the goals of CSE 262 is to teach you how to conduct research. A research project is a formal requirement of the course and counts for 60% of your grade. A project may be done individually, but some projects are better suited to a team of 2. You should allow 5 weeks to complete the project. Your project grade will be based on your successful completion of the following 4 parts:

1. Project proposal: due Thursday, April 28, 2011 (writeup + brief in-class presentation, 1/10 of the project grade)
2. Progress report: Tuesday, May 17, 2011, in-class (10-15 minute in class presentation, 1/10 of the project grade) Note date change
3. Final Project presentation: Presentation Schedule TBA
   Bring hard copies of your slides to class. (2/10 of your project grade)
4. Final Report: due Friday, June 3, 2011, 5PM, in EBU3B 3244 (Hard copy and electronic, 6/10 of the project grade)

A list of projects appears below. If you have your own project in mind, discuss it with me first.

Your project proposal should identify the investigator(s) and the title of the project. You should present a preliminary list of milestones—with completion dates—and possible options depending on your progress. Your milestones are not binding. If you need to fine-tune your goals as the project progresses, be sure to document your decisions. If you are working in a team, describe the division of labor in your proposal.

If you are proposing your own project, include a description of the project that thoroughly explains the bottlenecks of the application, the existing code base that you will use as a performance basis.

Your project proposal should address the following issues.

1. What are the goals of your project? Are they realistic?
2. What are your hypotheses?
3. What is your experimental method?
4. What result(s) do you want to demonstrate?
5. What are the significance of the results?
6. What code will you implement? Briefly describe any software packages or previously written software that you will be relying on.

The progress report will help keep you on a steady pace and enable you to make mid-course corrections. The report consists of a 7-10 minute presentation, and should address the following.

1. Discuss your progress to date. Complete the self-evaluation form. If working as a team, discuss the division of labor and other aspects of how you worked together as a team. If working on your own, answer the questions on the form that are specific to individual projects.
   A copy of the form is available in html.
2. Revise any milestones and completion dates that you established previously, including an explanation.

In evaluating your proposal and progress report I will check to see if your goals are realistic. A realistic project that is completed successfully will likely receive a higher grade than one that is too ambitious and doesn't get finished. Results alone are not adequate. You need to interpret them. If you have any questions about this, be sure and see me. If you do have an ambitious design, we can work together to set realistic goals.

Your report should be a polished piece of work. Read over these these requirements for your project which discusses what to present, what to include in your report, and instructions for how to submit an electronic copy). To get an idea of what I expect, look at project reports from courses I’ve taught previously.

Submit hard copy as well as an electronic version, e.g., html with ASCII and other attachments. Be sure to include an up-to-date Team Evaluation form.

Projects

Here is a tentative list of projects, but if you have a project of your own that you'd like to work on, let me know. Most of these can be formulated as one or two investigator projects. I’ll be adding to this list from time to time, so check regularly.

• MPI implementation of 3D multigrid. Reference material include
  • Multigrid tutorial slides by William Briggs
  • MG NET, a repository for information related to multigrid, multilevel, multiscale, aggregation, defect correction, and domain decomposition methods.
• Communication avoiding linear algebra algorithm in CUDA or MPI. For references: visit the Berkeley Bebop project web site: http://bebop.cs.berkeley.edu/#pubs
• Perform a performance qualification of Fermi, and compare with published results for the 200 Series GPUs. For example, study cache behavior. (There is a cache benchmark available.) Implement Matrix Multiplication, consider previous results described in V. Volkov and J. Demmel, Benchmarking GPUs to tune dense linear algebra, Proc. 2008 ACM/IEEE Conf. on Supercomputing. There are many other applications, consult these GPU resources for your options.

Maintained by [Thu May 19 22:02:57 PDT 2011]