Winter 2024: Cloud Networks

Instructor: Qizhen Zhang
Time: Friday 9:00 AM - 11:00 AM
Location: Zoom only (see the link on Quercus)
Discussions: Piazza
Office hours: Wednesday 9:00 AM - 10:00 AM or by appointment, Zoom only (see the link on Quercus)

Course Description
Cloud computing underpins today's largest online services. Enabling cloud computing are hyperscale data centers where data is stored, processed, and moved at unprecedented scales. Networks in and between data centers are one of the most fundamental components in the cloud---they connect millions of servers to provide computing as a utility.
In this seminar course, we first introduce cloud networks and then delve into the design principles and cutting-edge advancements of these networks. Specifically, we will discuss classic papers that have stood the test of time and recent advances published in top networking conferences.

Prerequisites
General background in computer systems and networking (e.g., CSC458/2209) is recommended.

Format
This seminar course consists of paper discussions and a final project.

• Paper discussions. We meet for two hours every Friday to discuss a classic and a recent paper on the same topic (see the schedule below). Before Friday 12 AM, students need to read both papers and submit the reviews (i.e., summary, strengths, weaknesses, and detailed comments on each paper) on HotCrp. During the class, all students are expected to participate in the discussion.

  • To drive the discussion, starting from the third class, each paper is presented by two students: an advocator and then an adversary.
  • The student who serves as the advocator introduces the work in the first 10 min, followed by 20 min showing its significance (e.g., strong motivation, innovative techniques, or the high impact it has generated or will generate) and addressing potential concerns.
  • The other student serves as the adversary and spends the next 20 min criticizing the work by discussing, for example, the unrealistic assumptions, potential deployment issues, or limited impact.
  • Other students can join the discussion anytime and will vote for a side at the end of the session. A class is divided evenly between two paper sessions.
  • Every student leads the discussion twofour times throughout the term.

• The final project. The second part of the course is to execute a research project on a specific topic related to cloud networking. Students first form teams of 2-3, each working on a one-page project proposal (plus unlimited references) that will be reviewed and approved by the instructor. At the end of the term, each team will give a 25-minute presentation (including 5-minute Q&A) and write a 6-page report (plus unlimited references).

  • This year's topic is shuffle, a communication primitive adopted by various cloud systems, e.g., MapReduce, databases, graph processing, and machine learning. Students can choose any perspective of shuffling in cloud data centers.
  • The project can be either the reproduction of a recently published work (or multiple of them) or the development of original ideas.
  • In the former case, students should understand, implement, and eventually improve the ideas of the work. Project quality will thus be evaluated by comprehension (20% if the selected work is clearly presented), completion (60% if the core ideas of the selected work are implemented and evaluated), and innovation (20% if new ideas are proposed, implemented, and evaluated).
  • If a team decide to focus on their own ideas, project quality will be evaluated simply by innovation. In that case, students need to justify the novelty by elaborating the design and comparing it to prior work (50%) and demonstrate the feasibility by implementing and evaluating the ideas (50%).
  • The project proposal should clearly state which approach the team pursues.
  • Experimentation platform: CloudLab (request to join the csc2229 project).

Grading

• Reading and participation: 50%
  • Paper reviewing: 15%
  • Leading discussion: 20%
  • Participation: 15%
• Final project: 50%
  • Proposal: 10%
  • Presentation: 20%
  • Report: 20%

Late Policy

• Paper reviews: 20% of the mark is deducted for late reviews before the class. No reviews will be accepted once the class is started. Students are allowed to skip the reviews of two papers without affecting their grade.
• Final project proposal and report: 20% of the mark is deducted for each day after the deadline.

Tentative Schedule