CSE708: Programming Massively Parallel Computers

This course is offered remotely via Zoom. Attendance is required. Please see the LMS (UBLearns) to find the Zoom information. The course is listed as an HE course, so will satisfy the requirement for a course to be taken on-campus/in person.

There is no prerequisite for this course.

The focus of this course is on experimental (hands-on) parallel computing. Each student is responsible for a semester-long project of their choosing, with approval from Prof. Miller. Grading (S/U) will be based on performance with respect to the project, as well as two talks (mid-semester and end-semester) that cover the project. Such presentations will include a definition and motivation of the problem, abstract sequential and parallel solution strategies, and an implementation of said solutions coupled with running times on parallel computers in the Center for Computational Research.

Your data will show results of strong and weak scaling. That is, of scaling based on Amdahl's law and Gustafson's observation. This will be discussed in detail in class.

• The first talk will provide a brief explanation of the proposed project, goals, expectations, and a timeline of the work to be performed.
• The second talk provides a summary of results and accomplishments.

Students are encouraged to look at the final talks from previous semesters' seminars and CSE633s classes.

Note that a successfully completed project satisfies the requirement for a project in the M.S. program. (The student who completes the project successfully is responsible for filling out the proper paperwork and presenting it to Dr. Miller for a signature.)

There will be a cap on the number of students allowed to enroll in the course, so that those who are enrolled will have a full experience and educational opportunity.

• For students who are new to parallell computing, you will write your programs in a standard programming language and use the MPI libraries for communication between processors.
• For those who have a good command of MPI, you may choose to use OpenMP or CUDA, if you wish. For those who are working with large-scale systems for the first time, it is typical that such students will use MPI. Please discuss your choice of a platform either in class or with Prof Miller.

All Seminars in CSE are graded S/U. Grading is subjective, based on the quality of the following:

• Class Attendance and Participation
• Solution strategy chosen that provides for a demonstration of strong and weak scaling, as discussed in class (ability to demonstrate both speedup and scaled speedup)
• Midterm Presentation
• Final Presentation

Presentations: