COS-518 is a graduate course in computer systems. Its goals are:
This course particularly focuses on the reading and study of recent research papers, with a heavy focus on distributed systems. We will discuss these papers in a small group in the course precepts.
The course includes introductory lectures to related topics. These lectures are joint with COS 418, see the lecture schedule here
The course also includes a semester-long implementation-heavy group project. COS-518 assumes a basic familiarity with computer systems and networking concepts.
A major component of this course is reading and discussing systems papers in the weekly precepts. We are only discussing 1 paper per week so we can discuss them in depth. You should closely read each paper and come prepared with several points that will substantially contribute to the group discussion.
General tips on reading system papers are here.
Your precept participation grade will be determined based on attendance and, more importantly, substantial contributions to paper discussions as tracked by our TA.
What we expect you to know and prepare before each discussion is here.
There will be a midterm and final based on lecture content that will be given jointly with COS 418. Once the midterm and final are scheduled they will be posted on the schedule
The semester-long project in COS518 is an systems-building project. Projects should be done in groups (either of size two or three, to be determined by the instructor after the course size finalized) and must involve significant system programming. All group students are expected to share equally in the implementation.
There is a significant departure this year in the scope of the project compared to most previous versions of COS518. In particular, we expect most students to satisfy the group project by reimplementing and reproducing the results from a paper we read during the semester or on a like topic.
As in earlier years, students are also able to satisfy the project by performing novel research projects, but such projects must be closely related to the material and topics taught in the COS518 curricula. The instructors will take a narrow view as to what projects satisfy this criteria (e.g., your ongoing research on software-defined networks will likely not be approved).~ ngoing research on self-driving cars or software-defined networks will likely not be approved).
Students uncertain as to the satisfiability of their project are urged to speak to instructors as soon as possible, as the initial timeline for project selection is aggressive.
There are two types of projects in this class: reproducing others’ research results, and novel research. Proposals should be submitted via a private note to Instructors on Piazza.
For reproducing research projects, students should write a note (a public one, so that your teammates can see/join in the discussion). The should be a few paragraphs that include the following information. Please tag these proposals as #proposal #reproduction.
As the final plan mentions, it is important when reading a paper to ask what questions and/or settings are not included in an evaluation. For example, what happens if a workload shifts from being uniformly distributed to Zipfian? What if failures occur in a different fashion than evaluated? What if the data in a big data processing system has a different structure than evaluated (e.g., the "graph" that the data represents has a different edge distribution)? And so forth…
For novel research projects, students should write a note (a public one, so that your teammates can see/join in the discussion). The should be a few paragraphs that include the following information. Please tag these proposals as #proposal #novel.
If students are concerned that their proposal project might not be sufficiently relevant to COS 518 to satisfy the topical criteria, please contact instructors earlier than later. Proposals not closely related to the topical matter may be rejected outright as not appropriate.
The project proposal from above will be finalized. This may involve one or more back-and-forth between instructors and group (likely via Piazza).
Rather than a research paper written as a Latex document, your final report will be a blog post that expands on a similar organization and topics as addressed in your initial project proposal.
For reproduction projects, the post should have a particular focus and discussion on the evaluation (setup, comparative results, discussion of differences, and so forth).
Your final report will be published on Medium as part of a course page. Source code and data should also be made available.