• Fall 2019 Syllabus
• First assignment Group photos

---

• Winter 2019 site

---

• My faculty page
• Free Hamilton C shell download for UMich students and faculty

Instructor

Nicole Hamilton
Lecturer III
Computer Science and Engineering
University of Michigan
2649 Beyster
2260 Hayward Street
Ann Arbor, MI 48109-2121

C: 425-765-9574
nham@umich.edu
https://web.eecs.umich.edu/~nham

Bio

I've spent most of my career as a system designer on new product teams and as an entrepreneur selling Hamilton C shell, a product I wrote in my living room 30 years ago.

In early 2003, I became the ninth dev on the team that created what became the Bing engine. I wrote the ranker, the part that orders the results, and the query compiler, that part that decides what the user is looking for. Overall, I contributed about 30 KLOC to the first release, representing about 10% of all the code in the backend as of the time we went live.

I never expected to be teaching at a university. I feel lucky to be here.

EECS 398 System Design of a Search Engine
Fall 2019

The plan is for this class to return next year, probably in W21, with the permanent number EECS 440.

Lectures
EECS 1200, Tuesdays and Thursdays, 3:00 pm to 5:00 pm
Lectures will be recorded but attendance is strongly advised.

Office hours
Beyster 2649
Mondays and Wednesdays, 1:30 pm to 2:30 pm
Tuesdays and Thursdays, 5:00 pm to 6:00 pm

Prerequisites
EECS 280 and 281

Classification
Depending on your major:
CE:  MDE or EECS elective
CSE:  MDE or Flex Tech elective
CS-LSA:  Capstone or Flex Tech elective
EE and DS:  Flex Tech elective

Overview

This is a course in how to tackle a large system programming project in C++. You’ll work on a team of 6 to write a complete multithreaded internet search engine from scratch.

My objective is to offer students an opportunity to work on a significant relatable project (something you can explain to your family and to recruiters) early in your undergraduate careers. I want the experience to be like working on the startup dev team for a new product (like the one I had on the startup team for what became the Bing engine.)

Sixty percent of your grade will be on the project and will be competitive, based mostly on your team’s performance compared to the others and on your individual contribution to your team. Thirty percent of your grade will be on the exams. Ten percent is homework.

System design

System design projects always seem to have some defining characteristics.

1. There’s an important domain-specific part that asks you to learn something new about an interesting problem you’ve never seen before, in this case, how a search engine works.

2. There’s a need to invent a solution, an architecture, breaking the problem down into lots of moving parts, with lots of data structures and algorithms.

3. It’s usually “close to the metal” with lots of low-level OS calls everywhere and the need to define file and interchange formats, perform handshakes, share resources using locks, deal seriously with error recovery, etc.

4. They're usually team efforts because they're too big to do any other way.

5. You build the whole thing, often from scratch, and at the end, you get to see it work and it feels good.

Every new design project always has a new domain-specific part, which keeps a career in system design interesting, but the rest of the skills are the same and they’re things most people learn by doing.

What I like about a search engine as a case study in system design is that it’s really compact fun project that hits on every bit of what, to me, system design is all about.

What to expect

This will be the third time for this as an experimental course. Each time, I get a chance to make it a little better.

I intend for the workload to be consistent with any other 4-unit EECS course and students report on the evals that that tends to be true. Because so much of this class is about working on a small team, consider signing up with some friends you'd like work with.

Engines typically run 6 to 10 KLOC (thousand lines of code). A few teams will decide to crush it with very sophisticated engines running 14 to 20 KLOC. The first time, most indices were pretty small, perhaps a few a few thousand documents. W19, most were over a million documents. One team hit 100M but accidentally DOS'ed the Duke Law School website during registration in the process. I admit feeling a little proud that I've unleashed such power in the hands of my students but dealing with the complaints, not so much.

The first time around, the big risk was whether an entire engine could be scaled down into a project that a small team of undergraduates could finish in one semester, especially as I'd never tried this before and I was starting with zero course content. But (tada!) every team successfully built and demonstrated a working engine. It needs to be said: Michigan students are amazing.