CSCE 590 Parallel I/O Systems

Course Requirements

Reading: We will do a lot of reading in the course! You should read the paper before the class period during which it is presented. Be prepared to discuss and share ideas about the paper.
Presentations: Each student will present at least two research papers from the literature. Each presentation is worth 5% of the grade. Presentations will be selected from papers in the text, or from other research sources as approved. If you would like you may work with another student on your presentation, but if you do then the presentation only counts for 1/2 credit. Two students working together must do two presentations for it to count as a single presentation for both students.

The schedule of presentations will follow the chapter order from the text. If a paper is selected that is not from the text, then it will be presented first in the set of papers from the chapter in which it most closely fits. We will schedule the various presentations on Wednesday, August 28, and Friday, August 30. The first student presentation will be on Wednesday, September 4.

As a part of the presentation, each student is required to provide three possible test questions on the paper. You do not need to provide written answers, although we may discuss the questions during the presentation, and you may be asked to provide written answers at some time in the future. These questions will become candidates for questions for the in-class final exam.

Presentations will be graded on the basis of how well you bring out the main points of the paper and how well you convey your understanding of the paper. A formal Power Point presentation is not required, but many students are more comfortable with this. You should plan to talk about 35 to 40 minutes and leave about 10 to 15 minutes to discuss the test questions you have prepared. Plan to be interrupted and have discussion during your presentation. The presentation is intended to enhance your education, not intimidate you. Generally, students receive a very high score on the presentation.

Final exam: 30% of the grade is based on the final exam. This is planned to be given at the end of the semester during the final exam period, on Saturday, December 14, from 12:30 to 2:30. The questions on this exam will be derived from the questions provided by students during their presentations, with modification and supplements by me.

Term paper: 30% of the grade is based on the term paper. This paper will be due near the middle of the semester, on Friday, October 4. I would like for you to write a 10 to 12 page 12-point double-spaced paper that compares two or more approaches to a particular problem that we have read about in class, or is a position paper that describes your approach to a particular problem. You must use at least three references, and the references can be papers from the text. You may use more references and outside references if you desire. All references used in the paper must be cited appropriately. Your paper must not consist only of statements from your references, but should include your insight and judgement about the paper, and should be in your own words. You may use one additional page for references. References may be single-spaced. Any paper longer than 15 pages of text plus one page of references will not be graded.

The paper you write should be an in-depth critique of the approaches in our literature, and should show that you have a very good understanding of the topic you have chosen. The paper should use good writing style. English and grammar will count as a part of the grade, so if you have difficulty with written English you should plan to start early and get some help from the writing lab here on campus.

As an example of a topic you might select, you could write on caching and prefetching techniques. For this topic you might compare and contrast the performance and ease of two or three of the caching policies described in chapter IV of the text. This is only an example and many other possible topics exist.

Grading on the paper will be competitive with the other students in the course. The quality of your paper will be compared with the quality of the paper of other students in the course. Better papers will receive better grades. Grading on the paper will be based on:

  • English and grammar: 20%
  • Choice of topic (is it relevant to the course?): 10%
  • Completeness, correctness, appropriateness, and sufficiency of references for the statements you make in your paper: 10%
  • Technical writing style (sections, references, bibliography, etc.): 10%
  • Depth of discussion or position (i.e., how many different points do you bring out, how well do you describe or defend them, are there any significant topics missing): 50%

This paper is to be done on your own. Obviously, any evidence of cheating on this assignment will be dealt with severely.

Semester Project: Project expectations can be found here.

30% of your grade will be on a semester project of your own design. I would like for you to consider some kind of programming project using a parallel I/O library such as MPI I/O, but various options are possible. We will have access to several types of cluster platforms for the project. The semester project will be due just before Thanksgiving on Monday, November 25. It is possible to work in teams of two. Some possible projects include:

  • Modify the disk simulator by David Kotz to simulate the performance of a modern disk system under various hypothetical workload or configuration scenarios. This simulator and others, including the modification of this one that was used in paper [15], are available at http://www.cs.dartmouth.edu/~dfk/diskmodel.html.
  • Modify a parallel application to use parallel I/O libraries. Perform various performance tests under a variety of conditions to test and report the improvement due to parallelism.
  • Write a *very* basic parallel file system.
  • Compare the performance and/or usability of two different parallel I/O libraries.
  • Examine, measure, and possibly improve inefficiencies in a parallel library or file system.
  • Starting with an existing parallel library or file system, design an extension of it to include a new caching or prefetching policy.
Grading on the project will be competitive with the other students in the course. The quality of your project will be compared with the quality of the project of other students in the course. I find that, generally, graduate students prepare very good projects.

Your project must be approved by me. I would like for you to turn in a one-page project proposal no later than Friday, October 11. The project proposal should state who is working on the project, the goals of what you hope to learn and accomplish by doing this project, and the general scope of the project. I will help guide you so that the scope of the projects in the course are similar to each other.

Important dates:
  • Paper scheduling day: Wednesday, August 28.
  • Term paper due: Friday, October 4.
  • Project proposal due: Friday, October 11.
  • Project due: Monday, November 25.
  • Final exam period: Saturday, December 14, 12:30-2:30.