Lecture and section information
CS 5220, Fall 2015
Lecture time: TR 8:40-9:55
Lecture location: Phillips 101
Staff and office hours
Prof: David Bindel
425 Gates Hall
TA: Colin Ponce
4 credits. Models for parallel programming and survey of parallel machines. Existing parallel programming languages, vectorizing compilers, and parallel libraries and toolboxes.
This class is intended for a broad array of students working with high performance computation, including students who might not have had a CS undergraduate degree. If you are concerned about whether you have the right background, please talk to the instructors. Generally, the prerequisites are:
- You will need programming skills in a C family language (C, C#, C++, Java). Code for the class will be written mostly in C (or C++).
- Prior exposure to a Unix environment is helpful but not critical.
- Prior exposure to numerical methods is helpful but not critical.
None required, but relevant texts include:
- Introduction to High Performance Computing for Scientists and Engineers (Hager and Wellein)
- Structured Parallel Programming (McCool, Reinders, Robison)
- Programming Massively Parallel Processors (Kirk and Hwu)
- Performance Optimization of Numerically Intensive Codes (Goedecker)
- Principles of Parallel Programming (Lin and Snyder)
- Introduction to Parallel Programming (Pacheco)
- Parallel Programming (Wilkinson and Allen)
- We will be collaborating on code and modeling exercises in class, and you are encouraged to bring a laptop to fully participate.
- You will need a free GitHub account. The Student Developer Pack option comes with various perks (including a micro account with three free repositories).
The primary computing platform will be an instructional cluster with fifteen Xeon Phi 5110P boards hosted in eight 12-core nodes consisting of Intel Xeon E5-2620 v3. Unless otherwise stated, homework and projects should be tested and timed on these machines.
Intel generously donated the Xeon Phi boards and funded the purchase of the host machines, which were provided with deep matching discounts by Dell.
Class material will be posted on GitHub, and any merged pull requests (corrections or additions to course notes, better solutions than ours, tests that uncover interesting problems, suggested exercises, etc) will be given extra credit.
Student work will also be posted on GitHub in public repositories, with homework and projects submitted via pull requests (don’t worry, we will discuss how these work in class).
CS 5220 will be run as a flipped classroom. You are responsible for reviewing assigned materials outside class; class time will be devoted to discussion of papers, pair programming, performance modeling exercises, and other active learning work. While it is okay to miss class sometimes, you are expected to participate regularly as part of your grade.
Readings will be assigned before class, and may include papers, notes, slide decks. There may also be occassional videos or interactive tutorials (even though these technically may involve little reading). We will assign a variety of pre-class responses to readings; these should be submitted prior to class, but may (and should) be updated after class in response to what was learned.
Homework and Projects
There will be a few small individual programming and performance modeling exercises to be done out of class (often with an in-class performance). There will be three (probably) standardized projects on performance analysis and tuning of example numerical codes, including both coding work and a written report. Project groups will be assigned, and students will be responsible for assessing the contributions made by each team member. The final projects are also in small groups, and may either involve a research project (default) or an instructional module (experimental):
- Research projects should examine performance analysis in some form, but are otherwise flexible. Examples might include exploring novel programming languages or models, enhancing or tuning a high-performance numerical code, or modeling performance tradeoffs in some existing code base. The deliverables for research projects will be a paper and a short presentation.
- Instructional modules should teach some aspect of high performance computing or related technologies, targeted at K-12, undergraduate, or graduate students. Examples might include a short course, a workshop, or a project for a course like CS 5220. We particularly encourage modules that use the Xeon Phi in some interesting way. The deliverables for instructional projects include not only the materials used (lesson plans, slide decks, sample codes) but also a set of learning objectives, an assessment plan, and a concrete plan for how the module might be deployed (though actual deployment is not required).
Class projects will be developed in public repositories on GitHub. Students are encouraged to study code written by other groups for inspiration (with citation), and to help provide other groups with constructive feedback. Project groups will also be responsible for reviewing the code and write-up of other groups under a rubric that will be provided.
Computations and performance experiments that appear in homeworks and project reports should be automatically reproducible on the class cluster, either by the instructors or by peers. We will penalize experimental results that do not come with an associated script that can reproduce them, even if they appear to be otherwise correct.
- Class participation and individual homework (20%)
- Standard group projects and peer review (50%)
- Final project (30%)
Late work policy
Pre-class homework is due at the start of class in order to ensure high-quality in-class exercises. Homework may be revised during or after class based on what is learned during the class. Grades for homework exercises will be based on a check/no check assessment of the initial submission and (potentially) a more thorough evaluation of a revised submission, which will typically be due before the next class.
Group projects are due by 11:59 on the due date in order to allow a proper opportunity for peer review. As with homework, group projects will be evaluated both before and after peer review, with a more thorough assessment after the review. Group projects that are submitted late will forfeit both the part of the grade associated with the initial submission and will not receive the benefit of peer review.
An assignment is an academic document, like a journal article. When you turn it in, you are claiming everything in it is your original work, unless you cite a source for it.
As part of the process of learning development skills, you should attempt to develop and debug code for yourself. Unlike most CS courses, we will by default keep all code in public repositories, and you are encouraged to look at code and writeups from other students for inspiration, as well as any resources you find on the web, provided you cite the work that inspired you. However, we do not encourage blatant bulk copying, even with citation. Such behavior will be visible not only to instructors but also to peers, and peer evaluators will be asked to identify such behavior.
We expect academic integrity from everyone. School is stressful, and you may feel pressure from your coursework or other factors, but that is no reason for dishonesty! By default, all work will be public on GitHub, and we encourage you to be inspired by work of others. But cite, and be aware of your own learning experience! If you feel you can’t complete the work on the own, come talk to the professor, the TA, or your advisor, and we can help you figure out what to do.
For more information, see Cornell’s Code of Academic Integrity.
Code of conduct
We have a code of conduct for contributing to the class (adapted from the Contributor Covenant 1.2.0). In addition to not harrassing each other, note that you must not publish private information without permission. Particularly when working on public collaborations such as group projects and peer review, please know and respect the privacy concerns of your peers. In public forums such as GitHub pull discussions, it is probably best to refer to each other only by GitHub identifiers.
GitHub and copyright
Students retain copyright for their academic work. While you will be asked to post code on GitHub, you are not required to post work under an open source license.
Any code or documentation you post for inclusion in the main class repository should be released under an appropriate open license; see the contributing guidelines.
GitHub and privacy
You will be asked to post your code on GitHub in a public repository, but this is not required. If you do choose to post code on GitHub, you are free to use a pseudonymous account. We will use the CS department Course Management System (CMS) to inform you of grades and evaluation of your work; this information will never be posted in public.
If you prefer not to use GitHub, this will not affect your grade in
any way. You will still use git to submit your code, but with
submissions created via
git format-patch and uploaded to CMS.
Your submissions will still be subject to peer review. If you want
to take this option, let us know promptly.
In the event of a major campus emergency, course requirements, deadlines, and grading percentages are subject to changes that may be necessitated by a revised semester calendar or other circumstances. Any such announcements will be posted to the course home page.