CSE 561: Computer Communication and Networks

Autumn 2026 · Mon/Wed 3:00–4:20 pm · CSE2 271 (Gates Center)

Overview

Computer networks are simultaneously a key enabler and a key bottleneck for most modern technologies that we rely on every day. No networks, no Slack, no Zoom, no TikTok; no networks, no online shopping, no photo sharing, no YouTube; no networks, no LLMs, no coding agents, no Google Maps ... you get the idea. There are almost 30 billion network devices on the Internet, that is, over 3 per person on this planet.

In this class, we will study how real-world networks are built, optimized, and operated. We will learn about the fundamental design principles and the state of the art for large networks, through classic and recent papers, talks by practitioners who run some of the largest networks in the world, and a research project of your own.

After this class, you will be better prepared to do research and development on computer networks, and on the distributed applications that depend on them.

Organization

The class is organized around reading and discussing papers, talks by leading practitioners, and a research project of your choice.

Readings	We will read classic and recent research papers, deeply reading 1-2 paper(s) for each class and shallowly reading 1-2 more. More on this below.
Discussions	We will have discussions online and in-class. The papers are a jumping off point; we will aim to have a broader discussion on the topic. For online discussions, you should finish all readings and post questions and comments on the discussion board before noon on the day of the class. You are also expected to actively engage in-class discussions. You will lead the discussion for one class in groups of 1-2. Sign up by sending your merge request to the course website
Guest lectures	We will have leading practitioners of industry sharing what they have been building. This is the best possible source for learning about the latest.
Project	You will do a research project since nothing beats doing it yourself. More on this below.
Grading	The project will account for 50% of your grade, participation in online discussions for 20%, leading the paper discussion for 20%, and in-class participation for 10%.
Textbook	There is no required textbook. If you are missing background for a topic, consider reading Peterson and Davie or consult the instructor.

As this is a discussion-oriented course, attendance in class is expected. There is the Zoom option for when in-person attendance is not possible (e.g., you are traveling or are sick). If you are not planning on attending the class in person, you should let the instructor know ahead of time.

Administrivia

Class	MW 3:00-4:20 PM PT Room: CSE2 271 in the Gates Center Zoom link: See Canvas
Course staff	Ratul Mahajan [ratul@cs.washington.edu] (instructor) Brody Franks [brof@cs.washington.edu] (TA) Office hours on demand.
Class mailing list	cse561a_au26@uw.edu Updated nightly from the registration list. Uses your UW (not CSE) email.
Canvas	canvas.uw.edu/courses/1931906
Ed board	edstem.org/us/courses/107537

Schedule

Date	Reading	Discussion lead, slides
Sep 30	Introduction Course overview and logistics.	RatulSlides
Oct 5	Philosophy Deep reads End-to-End Arguments in System Design The Design Philosophy of the DARPA Internet Protocols	Ratul
Oct 7	DC topology and network virtualization Deep read VL2: A Scalable and Flexible Data Center Network Shallow reads A Scalable, Commodity Data Center Network Architecture Network Virtualization in Multi-tenant Datacenters	Hongtao Zhang, Yuyao Wang
Oct 12	Guest lecture: Flat network topologies Deep read RNG: Flat Datacenter Networks at Scale Shallow reads Jellyfish: Networking Data Centers Randomly Xpander: Towards Optimal-Performance Datacenters A Throughput-Centric View of the Performance of Datacenter Topologies	Giacomo BernardiSr. Principal Scientist, AWS
Oct 14	Optical switching Deep read Beyond the Mega-Data Center: Networking Multi-Data Center Regions Shallow reads Realizing RotorNet: Toward Practical Microsecond-scale Optical Networking	Alice Zhu
Oct 19	Collective communication Deep read TACCL: Guiding Collective Algorithm Synthesis using Communication Sketches Shallow reads Demystifying NCCL: An In-depth Analysis of GPU Communication Protocols and Algorithms Due: Project proposal (2 pages)	Sheng Wei
Oct 21	ML networks Deep read RDMA over Ethernet for Distributed AI Training at Meta Scale Shallow reads An Introduction to the InfiniBand Architecture InfiniBand Credit-Based Link-Layer Flow-Control Credit-Based Flow Control for ATM Networks	Arhan Jain
Oct 26	End host stack Deep read Falcon: A Reliable, Low Latency Hardware Transport Shallow reads Snap: A Microkernel Approach to Host Networking
Oct 28	Guest lecture: Multipath Reliable Connection Deep read Resilient AI Supercomputer Networking using MRC and SRv6	Jitu PadhyeMember of Technical Staff, OpenAI
Nov 2	Global routing Deep read BGP Routing Policies in ISP Networks Shallow reads Understanding BGP Misconfiguration
Nov 4	Traffic engineering Deep read Engineering Egress with Edge Fabric: Steering Oceans of Content to the World Shallow reads Achieving High Utilization with Software-Driven WAN Decentralized Cloud Wide-Area Network Traffic Engineering with BlastShield
Nov 9	Network management Deep read Evolve or Die: High-Availability Design Principles Drawn from Google’s Network Infrastructure Shallow reads Pingmesh: A Large-Scale System for Data Center Network Latency Measurement and Analysis Due: Project progress update (2 pages)
Nov 11	Veterans Day – no class
Nov 16	Guest lecture: A whistlestop tour through AWS’s networking evolution Deep reads Jupiter Evolving: Transforming Google’s Datacenter Network via Optical Circuit Switches and Software-Defined Networking Datacenter Networks Are In My Way (James Hamilton, 2010 talk) Shallow reads AWS networking lab tour: Making networking disappear (The Register) A lot of datacenter networks are run by absolute clowns. Not Amazon’s (The Register)	Stephen CallaghanSr. Principal Technologist, AWS
Nov 18	Guest lecture: Networking from First Principles with Software at AWS Deep read Ravel: A Database-Defined Network Shallow reads Improving Network Availability with Protective ReRoute Layering as Optimization Decomposition: A Mathematical Theory of Network Architectures	Jorge VasquezSr. Principal Engineer, AWS
Nov 23	Guest lecture: AI in Physical Networks Deep read Intent-Driven Network Management with Multi-Agent LLMs: The Confucius Framework	Jamie GaudetteVP, Azure Global Networking
Nov 25	No class (day before Thanksgiving)
Nov 30	Guest lecture: Network Switches and Interconnects – The Lego blocks of Network Deep read Data Center Switch Architecture in the Age of Merchant Silicon	Satish VangalaDirector, Network Product Development, AWS
Dec 2	Guest lecture: Life as a packet slinger – lessons learned from several decades in the trenches Deep read Assessing the Aftermath: The Effects of a Global Takedown against DDoS-for-hire Services Shallow reads Click Trajectories: End-to-End Analysis of the Spam Value Chain Measuring the Cost of Cybercrime	Tom SchollVP and Distinguished Engineer, AWS
Dec 7	Guest lecture: Real-world network verification Deep read Lessons from the Evolution of the Batfish Configuration Analysis Tool Shallow reads Relational Network Verification	Dan HalperinPrincipal Scientist, AWS
Dec 9	Project presentations
Dec 172:30–4:20 pm	Project presentations (contd.) Due: Final report (6 pages)

Project

The project is a critical component of the class and is intended to provide you first-hand experience.

You should pick a topic of your choice. One option is extensions of your current research that are relevant to the topic of the class. Another is to use the aRPC framework [code, paper] to build a new type of protocol (e.g., for routing or transport) and benchmark its performance. Yet another is to develop an extension of a technique proposed in a recent paper. If you are unsure about the topic or scope, talk to the course staff.

The main deliverables for the project, due at the end of the quarter, are a short research report (6 pages) and an in-class presentation. There are two intermediate milestones as well. The milestones and their weight in the course grade are:

Oct 19: Project proposal (2 pages) – 10%
Nov 9: Progress update (2 pages) – 10%
Dec 9, 17: Presentation – 10%
Dec 17: Final report (6 pages) – 20%

You are expected to do the project in groups of 2 students. Talk to the instructor if you want to do it by yourself or in a larger group.

Reading papers

Reading and understanding academic papers takes some practice. If you are new to it, you may find How to Read a Paper helpful. In the terminology of this note, a deep read is three passes and a shallow read is the first two passes.

AI use policy

You are encouraged to use AI to deepen your understanding and to work more efficiently, but not in a way that circumvents the learning process. You may ask AI to explain parts of a paper or to provide background, but you must still read the papers yourself. You may ask AI to write code for your project, but you must be the designer and must intimately understand the details and failure modes of the system you build; you will be tested on that in the presentation. You must write the full first drafts of the project proposal, progress update, and final report yourself; you may use AI to polish them. In each deliverable, include information (prompts, process) on how you used AI.

You must not use AI to write your online discussion posts.

See this UW CSE policy for more details.