| Category | Difficulty |
|---|---|
| Reviews | 6 |
| Project | As hard as you want |
| Present | 6 |
This class gives an introduction to data center computing issues through reading research papers that address problems that datacenters face today, as well as working through a semester long research project about a data center problem.
This course allows you to get a holistic understanding of how the current hyper-scaled datacenter works in the industry across the technical stacks from hardware to software. It gives students an opportunity to glance into the challenges at the modern computing systems, and solutions that are deployed in the real-world systems. This course requires students to read one research paper before every class and conduct discussions during the class. The course also has a semester project where student can form team of 2-3.
The semester project will focus on one of the topics covered in the class. The course provides some project ideas and students are also free to proposed projects that align with the course material.
The structure of the class is as following: In general, each class will surround one paper to study. Three students will present to respectively give: summary, advocate, critics of the papers. Then the class will perform activities and discussion based on the paper. Guest speakers, usually one of the author or expert in related filed, from both the industry and academia will join at end of the class to answer questions students may have.
It is a very fun course if you like small class structure with discussions based lecture where you can get very in-depth insights.
If you are interseted in research, this would be a great course to get a sense of how paper reading is like and read a lot of great paper from both academia and the industry. If you are interested in the industrial, this is a great course to understand what are the real-world problem being tackled by all these big companies in their infra/system, eg: Google, Microsoft, Facebook, etc...
The course is taught by Professor Akshitha Sriraman. Here is the course website that can give an overview of the structure: https://github.com/STAR-Research/18-847B-F23/blob/main/README.md
Each class (meets twice a week), you are required to read and review one to two research papers. During class, these papers are presented and discussed. Sometimes, one of the researchers who wrote the paper will visit for discussion. The most important part of the reviews is the critical analysis of where they could have improved and your discussion questions, which are in your reviews. In general, the reading of paper not challenging once you get used to it. The expected time spent is ~3hr / paper. Student is required to read the paper and submit the summaries of the paper prior to class. Each student has 3 papers to skip throughout the semester.
You are required to prepare a set of paper presentations throughout the semester, ranging from 2-5 depending on class size. Student will get assigned to present a set of papers where you could be either be summary, avocate or critics of the paper. These presentations are 10 minutes long and can be relatively informal. Each class, two of these presentations occur, each one focusing on one of the papers for the class. These can take sometime to prepare, but was not meant to be long, contributing to a higher number of hours spent on the class for those weeks. Given that each paper contains so much information, condensing it into a shorter presentation can be difficult and take time to properly prepare.
Activities related to the paper will be conducted after the presentation. Some examples are: if you where to solve this problem at company X, what are the possible solution you'd design? Guest speakers are scheduled to join at the end of some classes to answer students' questions.
There is a semester long project for the class. It requires teams of three people and is meant to address something currently occurring in data center research. There are three milestones for the project. The first is before spring break, the second midway between spring break and the end of the semester, with the last milestone in the final week of classes. The project is not meant to be time intensive, however it does contribute the most to your final course grade. The professors and TAs are happy to help you come up with an interesting topic. Your team will have to write a report for each milestone to show progress. These reports are 1-2 pages in length and are meant to be more of a project update rather than an in-depth analysis of their system. However, the final report is more comprehensive.
You will walk through the plan with the professor to ensure the direction and viability. Students are expected to complete items being listed in the milstone that YOU proposed. But if it doesn't happen, it is also ok to talk about what you have tried and why it doesn't work out. Student will have meetings with the professor during the semester, and plans can be adjusted based on how the project evolve.
- I greatly enjoyed being exposed to new topics each week.
- Do not have to be familiar with every topic covered.
- Class discussions are very interesting and you learn a lot during them.
- TAs and professors are very approachable and easy to contact, since the class has a Slack workspace.
- Fast paced, difficult to ingest new information so frequently.
- Not enough time is given to the project in class.
- Reviews often take up a lot of time in the week.
- Attending in class is highly recommended. As participation and engagement does count towards your grade.
- Have a clear and viable plan for the semester project when creating the milestone plan.
- The course material covers many topics, don't be shy when you need to discuss material that you may not be familiar with. Creativity is highly encouraged during discussion. The point was not to design a perfect system in 15min, but rather demonstrate thought process.
- The course staff recommends picking a topics that you are excited about.
- Don't spend 10hours to read the paper and try to understand every details. The point was not to test you on how well you understand the details in the paper, but to understand the ideology and the approach researcher are taking towards the challenge.
-
Professor Akshitha Sriraman collects in-semester feedback during the course two times. It is a great way to raise any concern you may have with the course or propose ideas. Note: make sure you fill out the form because extensions are granted after you fill it out :)