- Course Instructor: Prof. Tian Guo
- Semester: Spring 2020
- Office: Fuller Labs 138
- Office Hours: Mondays 4:30 to 5:30pm; Thursdays 2:00 to 3:00pm
Course Description
Class: Mondays, 6:00pm to 8:50pm in Higgins Labs 154
Modern data centers, with hundreds of thousands of interconnected physical servers, provide the infrastructure foundations for today’s evolving cloud computing platforms. Cloud computing, with its clear economic benefits and flexible resource offerings, hosts a plethora of services ranging from traditional web services to deep learning jobs.
In this research seminar, we will discuss both industry and academic research on cloud computing. Our goals include understanding of both key technical challenges and existing solutions in today’s cloud platforms. Students will read two papers per class and engage in active discussions, as well as perform a semester-long project in groups of 2-3.
Grading
This course will be graded based on a project (50%), paper reviews (20%), paper presentations (15%), and participation (15%).
Class Format
Note: This course will use both InstructAssist and Google Drive for distributing and sharing course materials. You could expect to find most of the materials through three Google Drive directories. InstructAssist will be mainly used for submitting paper reviews, peer review, displaying course grades, and forum discussion.
Please fill out this short survey to specify your presentation and scribe preferences, before January 20. I will try to accommodate your preferences as much as I can and will post the presentation schedule in this google sheet.
If you are presenting, please put the Google Slides for your presentation in this folder before the corresponding class meet time. If you are assigned to scribe for a class, please take the notes in a Google Doc and add the notes to this folder right after the class. You could continue polishing the notes but you should aim to finish a final version before the next class.
It is critical to save the deliverables to the correct folders, and follow the file naming conventions of the example file in the folders. To obtain read/write permissions to the folders, please see the instruction here.
Schedule
All papers can be downloaded from here. The instructor might decide to replace the listed papers with better alternatives, as the course progresses. Such replacements will be announced at least two weeks ahead, both in class and through the forum.
| Date | Topic | Readings | Additional Resources |
|---|---|---|---|
| 1/15 | W1: Intro to Cloud | A View of Cloud Computing | |
| 1/20 | No Class (Martin Luther King Day) | ||
| 1/27 | W2: Edge and Serveless Computing | The Emergence of Edge Computing, Cloud Programming Simplified: A Berkeley View on Serverless Computing | |
| 2/3 | W3: Workload Study | Analysis of Large-Scale Multi-Tenant GPU Clusters for DNN Training Workloads, Usage patterns and the economics of the public cloud | Additional: google trace |
| 2/10 | W4: ML Serving | PRETZEL, MArk | Michelangelo and FBLearner Flow |
| 2/17 | W5: Monitoring and Analysis | Performance Monitoring and Root Cause, Performance Analysis of Cloud Applications | Production bugs |
| 2/24 | W6: Resource Management and Scheduling | Large-scale cluster management at Google with Borg, Apollo | Resource scheduling survey |
| 3/2 | W7: Resource Allocation and Provisioning | Resource Central, Hcloud | |
| 3/9 | Spring Break | ||
| 3/16 | W8: Cloud Storage | Dynamo, Aurora | |
| 3/23 | W9: DL Clusters | Optimus, Tiresias | |
| 3/30 | W10: Microservices | Service fabric, Seer | |
| 4/6 | W11: Economics | CherryPick, Tributary | PARIS, Proteus |
| 4/13 | W12: Serverless | SAND, Model-less serving | Occupy the Cloud: Distributed Computing for the 99% |
| 4/20 | No Class (Patriots’ Day) | ||
| 4/27 | W13: Project Presentation | ||
| 5/4 | W14: Project Presentation and Review |
Project
Students will propose and run a semester-long project, ideally in groups of 2-3. Please see here for a list of potential project ideas. It is fine to use your existing research project if it is relevant to the course and the instructor approves. You will present the project at the end of the course and write a 10-12 page report. If you are unsure about scoping the project, please visit my office hour.
Please refer to this document for more instructions on each milestone of the project.
Project Timeline
- Project proposal: 5PM on Friday Feb 14 (W4)
- Mid-term review: 5PM on Friday March 27 (W9)
- Presentation: 5PM on Monday April 27 (W13)
- Final report: 5PM on Monday May 4 (W14)
Paper Review
Every week, there will be two paper reviews due at 5PM of the Friday before class. All students except the week’s presenters should submit the reviews through here.
Review Grading
Throughout the semester, the instructor will randomly select and grade five reviews per student. The final review grade will be an average of the five review grades. Students may miss up to two reviews per semester without impacting the final grade. Each review will be graded on a 0-10 scale.
Peer Review
You will be provided with opportunity to practice peer review. After each paper review deadline, you will have access to two randomly assigned review submissions and use the online form to critique each other’s review. Please be professional and focus on suggestions for improvement. Peer review will be counted toward the participation score.
In-class Presentations
Students will specify their preferences and will be assigned one paper (maybe two) to present in class. The assigned presenter must prepare sufficient material for 45 minutes and submit the Google Slides of your presentation to the shared folder. You are encouraged to engage with the rest of the class by asking questions and soliciting opinions during your presentation.
You are expected to cover the assigned paper, but also can choose to include other relevant background information or papers within the same topic area. If you are unsure about what to be included in your presentation, please come by during the instructor’s office hour. You should allocate enough time to prepare this presentation as a good presentation can take up significant amount of time.
Obtain Access to Shared Directories
We will use shared Google Drive directories to share in-class presentation slides, class notes, as well as other resources. These directories are not available to public and will only become accessible once you successfully join the Google Group wpi525tcloudcomputing.
To join the Google Group wpi525tcloudcomputing, you will need to first sign in to the Google Group service; and then search for “wpi525tcloudcomputing” to apply for the membership. Once I approve your membership, you will get an email notification and will then be able to see the links to the shared Google Drive directories.
Academic Honesty
The WPI Academic Honesty Policy describes types of academic dishonesty and requirements in documentation. In the case of academic dishonesty, I am required to report the incident to the Dean of Student Affairs. Plagiarism will lead to a failing grade and may bring more serious consequences at WPI level.