CS735/CS835 Programming Project

Your semester project should involve the writing of a parallel program. In essence the project is an extended, student-designed, programming assignment. The effort expended should be roughly equal to 1.5 times the effort expended (on average) on the conventional programming assignments in this course.

There are three deadlines:

1. October 7: You must submit a detailed project plan.

   This plan should include introductory paragraphs on what the problem is to be solved, why you think it may lend itself to a parallel solution, and what you hope to learn by doing the project.

   The plan should also include a description of goals for an intermediate checkpoint (see next bullet item) and for the final submission. These goals must be very concrete and must include explicit criteria for measuring your success at accomplishing the goals. The goals should be specified in terms of awarding points for subgoals accomplished. The success criteria should be objective and clearly defined. Usually this will mean that some component of your project will be coded, tested and evaluated. This evaluation step is important. Be sure to give some thought as to how you will be evaluating the results of your program. Also consider how the quality of that evaluation should be judged.

   The plan should be typed. Electronic submission (on agate.cs.unh.edu) is preferred:

   ~cs735/bin/submit projplan file

   My submission mechanism will not handle large Postscript files so you may also submit hard copy in person.

   This hurdle is worth 20% of the project grade. Your grade will be determined by the educational value of your proposed project and the specificity of your plan. The plan may be turned in late at a penalty of 5% (of the 20%) per day up to one week late. Your plan, as amended by me when I grade your plan submission, will constitute the official plan under which you will complete the project.

   You are encouraged to discuss your ideas with me informally prior to submitting the formal plan.

2. November 11: You must submit a report on your status at the intermediate checkpoint. The submission should include all source code and the text for any discussion of the results. Any text should be typed and, at least, minimally formatted. Source code should be appropriately structured and documented.

   Most importantly, the report should evaluate your progress against the checkpoint goals specified in your (possibly amended) project plan. Give yourself an explicit grade by assigning points as indicated by your specified goals and criteria.

   This hurdle is worth 40% of the project grade. The checkpoint report may be turned in late at a penalty of 5% (of the 40%) per day up to one week late. The grading of your submission may include a face-to-face meeting with me outside of class. Students with particularly interesting results may be asked to give brief presentations in class as well.

   You should electronically submit all your source code. Please include a README file in your submission that describes what all the files are. Also include a Makefile for building your programs. Electronically submit your report as well, if possible. If not, give me hard copy in person.

   ~cs735/bin/submit checkpt files

3. December 9: You must submit the final project report.

   This is simply a repeat of the process used at the intermediate checkpoint.

   This hurdle is worth 40% of the project grade. The final report may be turned in late at a penalty of 5% (of the 40%) per day up to one week late. To turn in your files,

   ~cs735/bin/submit final files

   As before, hard copy submission of the text of your report is acceptable.

• Solve some problem you are interested in, using a parallel program on parallel hardware. Evaluate the performance of the program. This is the prototypical project.

• Re-do one of the three programming assignments using an alternative programming language. See the literature survey specification for possibilities. Evaluate the strengths and weaknesses of the two approaches. Also evaluate the performance results.

• Some people have claimed that the message-passing paradigm is best, even if you are executing on a shared-memory machine. Explore this claim by implementing the small message-passing library available on the Galaxy cluster for execution on agate. An evaluation should be done by recoding program two (linear system solver) to use threads and then comparing the threads version to the message-passing version on agate. Evaluate both performance and ease of programming.

• Implement a subset of the C Vector Library (CVL) using multithreading on agate. Implementations of CVL are available from Carnegie Mellon University and Dartmouth College. Evaluate the performance of your implementation by executing benchmark programs.

• Investigate the use of parallelism in data mining or data compression applications. For example, data from satellite instruments are being gathered in mind-numbing volume. The question remains as to how best to utilize all that data. Investigate, and do prototype implementations, of a data manipulation function that executes in parallel on agate. A data mining application might read through a collection of data searching for some "interesting" relationship among the data. A data compression application might use wavelets technology.

• Investigate the use of C* to do data mining or data compression applications.

• The data-parallel scan operation computes parallel prefixes. For example, scan with add computes prefix sums of a list of values. Implement a subset of the C* scan function using multithreading on agate or message passing on the Galaxy cluster. See the C* Programming Guide for the details of the C* scan operation.

• The data-parallel put operation allows processors to write values to the memory assigned to other processors. This operation can also have an associated combiner to resolve collisions. Implement the C* put function using multithreading on agate. See the C* Programming Guide for the details of the C* put operation.

• Implement a parallel constraint solver using multithreading on agate or message-passing on the Galaxy cluster. Evaluate the performance of your program.

• Port the Galley parallel file system to the Galaxy cluster.

• Port the Stream* data-parallel I/O library to the Galaxy cluster.

• Investigate the use of Java for parallel programming. Implement one of the programming assignments using Java threads.

• Port the Otter MATLAB compiler (by Mike Quinn) to the Galaxy cluster.

• Implement the shallow water benchmark using message passing on the Galaxy cluster. Source code for a serial version of the program is available in

      /home/cs/cs735/public/shallow/shallow0000.c
  

Comments and questions should be directed to pjh@cs.unh.edu