The course meets TR 2:30-3:55pm in Annenberg 213.
CS/EE 146 is a topics course in networking and this is the version is offered in Spring 2011. Our focus will be on energy-efficient designs of computer systems at all levels: including data centers, speed scaling, routing, storage, etc.
Adam Wierman, adamw@caltech.edu
Zhenhua Liu, zliu2@caltech.edu
This will be a research oriented course and the organization will be focused around reading papers. During each class we will plan to have a presentation and discussion of 2 related papers. A student will give the presentation and lead the discussion, though I'll help to promote discussion.
Every student is expected to have read the papers before the class. To enforce this each student will need to turn in 1/2-1 page summaries of the papers to be discussed by email to Adam before 1pm the day the papers are being presented.
This is a preliminary breakdown that may change during the term. Note that the class participation is a significant component of the course since the in-class discussions are fundamental to the success of the class. Attendance is mandatory, though valid excuses such as interview/conference travel and sickness are acceptable.
Grading of summaries: Each summary will get a check plus (5 pts), a check (4 pts), or a check minus (2 pts). Getting a check on every one is enough to get you an "A" for the summaries component of the grade.
Grading of presentations: At the end of each presentation all students will submit comments and grades for the speaker using this form. The grades for the presentations will then be a weighted combination of the grades from the students and Adam's grade for the presentation.
Nothing will be accepted late. I want people to have read the papers before the classes and I want to see your thoughts before the discussions, and none of this is accomplished if you turn things in late. There are enough paper summaries that if you miss a couple your grade won't suffer too much.
Raga & Desmond: The datacenter as a computer, Barroso & Holzle.
Rishi: The facebook data center FAQ
Riley: A look inside wikipedia's infrastructure and some other references are on James Hamilton's blog
Xiaofei: Computer architecture techniques for power-efficiency. Kaxiras and Martonosi. Chapters 1-3, 5.1, 6
Sasha: Intel speedstep white paper
Isaac: Power provisioning for a warehouse-sized computer. Fan, Weber, and Barroso.
Robert: PowerNap: Eliminating server idle power. Meisner, Gold, and Wenisch.
Changhong: Online Strategies for Dynamic Power Management in Systems with Multiple Power Saving States. Irani, Shukla and Gupta.
Masoud: Optimal power-down strategies. Augustine, Irani and Swamy.
Christina: Energy-efficient algorithms.
JK: A scheduling model for reduced CPU energy. Yao, Demers and Shenker.
Claudia: Optimality, fairness, and robustness in speed scaling designs. Andrew, Lin, and Wierman.
Jonathan: Speed scaling of processes with arbitrary speedup curves on a multiprocessor. Chan, Edmonds and Pruhs.
Fred: Load balancing and unbalancing for power and performance in cluster-based systems. Pinheiro, Bianchini, Carrera, and Heath.
Sonal: Energy aware server provisioning and load dispatching for connection-intensive internet services. Chen, He, Liu, Nath, Rigas, Xiao, and Zhao.
Desmond: Managing server energy and operational costs in hosting centers. Chen et al.
Lingwen: Dynamic right-sizing for power-proportional data centers. Lin, Wierman, Andrew, and Thereska.
Christina: When to Reap and When to Sow: Lowering Peak Usage With Realistic Batteries. Barnoy, Feng, Johnson, Liu.
Riley: Peak shaving through resource buffering. Barnoy, Johnson and Liu, 2008.
Kai: The cost of a cloud: research problems in data center networks. Greenberg et al.
Lingwen: A scalable, commodity data center network architecture. Al-Fares et al.
Masoud: VL2: A scalable and flexible data center network. Greenberg, Hamilton and Jain.
Rishi: ElasticTree: Saving energy in data center networks. Heller et al.
Jonathan: Making scheduling "cool": temperature-aware workload placement in data centers. Moore, Chase, Ranganathan, and Sharma.
Robert: Cool job allocation: measuring the power savings of placing jobs at cooling-efficient locations in the data center. Bash and Forman.
Zhenhua: DRAM errors in the wild: A Large-Scale Field Study. Schoreder, Pinheiro, and Weber.
Bose: Sierra: a power proportional, distributed storage system. Thereska, Donnelly, and Narayanan.
Changhong: Robust and flexible power-proportional storage. Amur, Cipar, Gupta, Granger, Kozuch, and Schwan.
Isaac: Cutting the electric bill for internet-scale systems. Qureshi, Weber, Balakrishnan, Guttag, and Maggs.
JK: Donar: Decentralized server selection for cloud services. Wendell, Jiang, Freedman, and Rexford.
Raga: Greening geographical load balancing. Liu, Lin Wierman, Low, and Andrew.
Sasha: Geographical load balancing with renewables. Liu, Lin, Wierman, Low, Andrew.
Claudia: Capping the brown energy consumption of internet services at low cost. Le, Bianchini, Nguyen, Bilgir, and Martonosi.
Fred: Greening the Switch. Ananthanarayanan and Katz.
Kai: Measuring Minimum Transmission Resources in the Internet Access Router: A Sustainable Performance/Energy Trade-Off Approach. Lombardo, Panarello, and Schembra.
Minghong: Reducing network energy consumption via sleeping and rate-adaptation. Nedevshci et al.
Bose: Greening Backbone Networks: Reducing Energy Consumption by Shutting Off Cables in Bundled Links. Fisher, Suchara, and Rexford.
Sonal: Greening the Optical Backbone Network: A Traffic Engineering Approach. Xia et al.
There is a lot of reading for this course and some of the more mathematical papers may require concepts that some of you haven't learned... But, that doesn't mean you shouldn't be able to get something out of each of the papers. The summaries are meant to ensure that everyone has, at least, read enough to understand the context and goals of the papers, and can actively participate in the discussions about the papers.
A good summary should include:
The summaries do not need to be longer than 1/2 - 1 page. So, I encourage you to be precise and to think critically about the papers. The papers are picked to provide either an introduction to an area or a starting point for our discussions. Hopefully, each section can leave us with a bunch of interesting research questions to pursue.
The job of the presenter is much more involved than what is needed to write the paper summaries. I expect that the presenter will not only read the paper assigned but a 2-3 other papers that are related (and maybe cited by the paper assigned). Often the papers in a class are very related, so it makes sense for the speakers to plan the presentations jointly, to some extent, to ensure that the same background is not covered multiple times.
The goal of the presenter should be to
When writing the presentation the speaker should feel free to use whatever slides they can find online to help with the presentation. Often authors will post slides from their talks about the papers. Also, there was a course organized by Kirk Pruhs where a number of the papers we will discuss were presented. Feel free to use such slides, but be sure to add your own insights and thoughts.
A typical talk should be ~30min and lead to discussion ~15min of discussion either during or afterwards.