Kick-off meeting: October 22nd, 09:00—10:00, LC 140
Weekly lecture: Tuesdays, 14:00—15:00, LF 035
Instructors:
Thomas Fogal,
Jens Krüger
Office: LE 305
Phone: 203 379 1314
E-Mail: thomas.fogal at uni-due punkt de
Office hours: Tuesdays, 15:00—16:00 (directly after class), or
by appointment (send an email).
NOTE: This course will be given in English!
Course Description:
Implementation-heavy course focused on performance-critical code. Parallelism models; messaging-passing and shared-memory architectures. Modern technologies for parallelization; OpenMP, MPI, CUDA. I/O performance issues; parallel and distributed filesystems. Network technologies in clustered environments. Deep storage hierarchies and the memory wall.
This is an implementation-intensive course.
There are no explicit prerequisites, but it is recommended you have a basic background in operating systems. You may find Scientific Visualization to be useful, but we will discuss any needed topics from that course in an ad hoc manner in this course. Please talk to the instructor if you have any doubts about your readiness.
This course has the following objectives:
If you plan to take the course, please register on the courses' moodle page. The registration code is just 'hpc' (all lowercase).
The implementation part of this course will be evaluated by multiple large programming projects. You will implement a simple `N-Body' simulation as well as analysis tools to understand the data output by your simulation. Later, you will apply the knowledge you have gained in a custom program based on your interests. Your solutions should be written in C, C++, or Fortran 90+. (Exceptions to using these languages may be allowed, but you will need to negotiate them with me.)
We may include or additionally offer a crash course in C, if there is enough interest.
An N-Body simulation is a simulation of the movement of bodies that
interact with one another. One application is (for example) computing
planetary motion according to Newtonian gravity. The Figures below give
2 examples of such a system.
Class time will be used to help guide you in the implementation of your N-Body simulation. The course will start out simple and progressively build more and more complex parallelism (and thus higher-performance!) into your simulation. In the end, you will have created your own highly-performant N-Body simulation that utilizes multi-scale parallelism.
The project is cumulative. You must live with your code for the whole semester. Solutions for phases will not be given out. Regression testing is critical.
Each project will be worth a major part of your grade. Exact point values will be discussed during the kick-off meeting. Assignments are due at midnight on the day they are due.
WARNING: All groups are expected to do their own work on the programming assignments (and exams for that matter). No cross-group collaboration is allowed. A general rule to follow is that you may discuss the programs with other groups at the concept level but never at the coding level. If you are at all unclear about this general rule, do not discuss the programs with other students at all.
There are no required textbooks for this course. However, you may find Viktor Eijkhout's HPC book useful for more depth into the concepts we cover. Once you understand the basics of C, you may find Axel-Tobias Schreiner's treatise on how to do object-oriented programming in C to be enlightening.
Other resources you may find useful are the documentation for:
The CCSS Cray-XT6m HPC cluster (duecray.uni-due.de) will be the primary computing resource for this course. Account creation/registration will be worked out on the first day of class. Your program will be graded on the Cray and so you must test your program in that environment! However, we recommend you setup a local machine or VM to do most development on, as the Cray is a shared resource.