Announcements

Remember to check this regularly!

• Starting Monday Jan. 14, we will be in Tech L221 to fit demand.
• Take the tour – http://tour.golang.org

Course Description

Distributed systems are collections of networked computers that coordinate their actions through message exchanges. Most computing systems you interact with everyday are indeed distributed (e.g. email, web, Google, Skype, Facebook ...) for a variety of reasons such as fault tolerance, performance, and the geographical nature of the requirements.

In this course, we will discuss some of the basic principles behind distributed systems and review some of the main paradigms used to organized them.

In compliance with Section 504 of the 1973 Rehabilitation Act and the Americans with Disabilities Act, Northwestern University is committed to providing equal access to all programming. Students with disabilities seeking accommodations are encouraged to contact the office of Services for Students with Disabilities (SSD) at +1 847 467-5530 or This email address is being protected from spambots. You need JavaScript enabled to view it. . SSD is located in the basement of Scott Hall. Additionally, I am available to discuss disability-related needs during office hours or by appointment.

Course Prerequisites

• EECS-343 Operating Systems (or equivalent)
• EECS-340 Introduction to Networking.
• (Please contact me if you do not meet some of these requirements.)

Communication Channels

There are a number of communication channels set up for this class:

• We will use the course web site (http://aqualab.cs.northwestern.edu/classes/eecs-345-w13) to post announcements related to the course. You should check this regularly for schedule changes, clarifications and corrections to assignments, and other course-related announcements.
• We will use the Google Group EECS 345 at Northwestern for class discussion. This particular channel is intended to foster communication among you, the students. You'll find that someone else in the class will have thought of the same problem that you have and will perhaps have some valuable insight that will prove helpful. The staff will be monitoring the discussion threads and will step in with guidance when appropriate. If you don't have a subscription already, you can request one:
  Email:
• There is always email for questions that would be inappropriate to post on the newsgroup/discussion-board. When using email to contact the staff please start your subject line with "eecs345: helpful-comment" to ensure a prompt response.

Course Organization

The course is organized as a series of lecture and paper discussions, two five-week projects and a take home exam.

• Lectures - A set of lectures on the core of the material.
• Readings - Textbook and paper reading in preparation for (not substitution of) the lecture.
• Projects - You will build at least one system that addresses issues, solves problems and exploits techniques from classroom discussions and readings.
• Exam - A take-home, final given in the last week of class.

Grading

I use a criterion-referenced method to assign your grade; in other words, your grade will be based on how well you do relative to predetermined performance levels, instead of in comparison with the rest of the class. Thus, if a test has 100 possible points, anyone with a score of 90 or greater will get an A, those with scores of 80 or greater will get a B, those with scores of 70 or greater will get a C, and so on. Notice that this means that if everyone works hard and gets >90, everyone gets an A.

Total scores (between 0 and 100) will be determined, roughly, as follows:

• Summaries of reading assignments 20%
• Class participation 20%
• Project 40%
• Exam 20%

Topics

• Introduction
• Networking and Internetworking
• Communication and overlay
• Naming
• Time and global state
• Coordination and agreement
• Concurrency control
• Consistency and replication
• Fault tolerance
• Protection, security and censorship

For each of the topics, we will first review introductory material before presenting and discussing a related research paper to help you better understand the main issues.

Remember to blog on your reading.

Week	Date	Topic
1	1/7	Introduction/system models/Go Defining distributed systems and their goals, hardware and software concepts, the client-server model and architecture, modern architectures for distributed systems. An overview of major systems models; we will use a couple of current papers to make our discussion more concrete. [Slides pdf] (This is good background material and, if you are a grad student in systems, you should probably read it) J. Gray, Distributed computing economics, Microsoft TR, 2003 J. Ousterhout, The Role of Distributed State, 1991 A short introduction to Go [Slides pdf] Take the tour – http://tour.golang.org
	1/9	Paper discussion (this is required reading and blogging) E. Brewer, Lessons from Giant-Scale Services, IEEE Internet Computing, July-August, 2001 Presenter: Zach Bischof [slides] M. Freedman, Experiences with CoralCDN: A Five-Year Operational View, In Proc. of NSDI, 2010 Presenter: John Rula [slides]
2	1/14	Networking and Internetworking (A quick review of networking topics) [Slides pdf] D. Clark, The design philosophy of the DARPA Internet Protocols, In Proc. of ACM SIGCOMM, August 1988 J.H. Saltzer, D.P. Reed, and David Clark, End-to-End Arguments in System Design, In Proc. of ICDCS, August 1981
	1/16	Paper discussion D.D. Clark and M.S. Blumenthal, Rethinking the Design of the Internet: End-to-end arguments vs. the brave new world, ACM Transactions on Internet Technology, 2000. Presenter: Josiah Matlack [slides] J. John, E. Katz-Basset, A. Krishnamurthy, T. Anderson and A. Venkataramani, Consensus Routing: The Internet as a Distributed System, In Proc. of NSDI, 2008 Presenter: David Tuber [slides]
3	1/21	Off - MLK
	1/23	Communication and overlays A review of common communication models including RPC, publish-subscribe and multicast. [Slides pdf] A. Birrell and D. Nelson, Implementing Remote Procedure Calls, ACM Transactions on Computer Systems, 2, 1, pgs. 39-59, February 1984. P. Eugster, P. Felber, R. Guerraoui and A-M. Kermarrec, The many faces of publish/subscribe, ACM Computing Surveys, 35(2), 2003.
4	1/28	Paper discussion Dabek et al, Designing a DHT for low latency and high throughput, In Proc. of NSDI, 2004 Summary: Yixi Zhang Presenter: Besim Avci [slides] M. Castro, P, Druschel, P. Kermarrec and A. Rowstron, SCRIBE: A large-scale and decentralized application-level multicast infrastructure, IEEE Journal on Selected Areas in Communication (JSAC), 20:8, 2002. Summary: Josiah Matlack Presenter: Byungjin Jun [slides]
	1/30	Naming A look at the naming problem and different naming models and systems. [Slides pdf] P. Mockapetris and K. Dunlap, Development of the Domain Name System, In Proc. of SIGCOMM 1988. H. Balakrishnan, K. Lakshminarayanan, S. Ratnasamy, S. Shenker, I. Stoica and M. Walfish, A layered naming architecture for the Internet, In Proc. of SIGCOMM, 2004.
5	2/4	Paper discussion J. Jung, E. Sit, H. Balakrishnan, and R. Morris, DNS performance and the effectiveness of caching, (fixed link) ACM SIGCOMM Internet Measurement Workshop 2001. Summary: TBD Presenter: Andrew Kahn [slides] V. Ramasubramanian, E. Sirer, The Design and Implementation of a Next Generation Name Service for the Internet, In Proc. of SIGCOMM 2004. Summary: TBD Presenter: Saurabh Kadekodi [slides]
	2/6	Time and global state Synchronization in distributed systems, logical time, global state. [Slides pdf] L. Lamport, Time clocks and the ordering of events in a distributed system, Communications of ACM, 1978 M. Raynal and M. Singhal, Logical time: Capturing causality in distributed systems, IEEE Computer, 29(2), February 1996 M. Chandy and L. Lamport, Distributed Snapshots: Determining Global States of Distributed Systems, ACM TOCS, Feb. 1985
6	2/11	Paper discussion X. Liu, Z. Guo, X. Wang, F. Chen, X. Lian, J. Tang, M. Wu, M Frans Kaashoek and Z. Zhang, D3S: Debugging Deployed Distributed Systems, In Proc. of USENIX NSDI, April 2009 Summary: Jacob Sherin Presenter: Donghan Miao [slides] M. Yabandeh, N. Knezevic, D. Kostic and V. Kuncak, CrystalBall: Predicting and preventing inconsistencies in deployed distributed systems, In Proc. of NSDI, 2009 Summary: TBD Presenter: Xiangyi Xie [slides]
	2/13	Concurrency control Mutual exclusion, election, communication. [Slides pdf] Ricart, Agrawala, An optimal algorithm for mutual exclusion in computer networks, CACM 24(1), 1981 Sanders, The information structure of distributed mutual exclusion algorithms, ACM TOCS 5(3), 1987
7	2/18	Paper discussion P. Hunt et al., Zookeeper: Wait-Free Coordination in Internet-scale Systems, USENIX ATC, 2010 Summary: TBD Presenter: Marcel Flores [slides] G. DeCandia, D. Hastorun, M. Jampani, G. Kakulapati, A. Lakshman, A. Pilchin, S. Sivasubramanian, P. Vosshall and W. Vogels, Dynamo: Amazon's Highly Available Key-Value Store, in the Proc. of SOSP, 2007. Summary: TBD Presenter: Byungjin Jun [slides]
	2/20	Coordination and agreement Consensus and related problems. [Slides pdf] Leslie Lamport, Paxos Made Simple, ACM SIGACT News (Distributed Computing Column), Volume 32, Number 4, December 2001 M.J. Fisher, N.A. Lynch and M.S. Paterson, Impossibility of distributed consensus with one faulty process, Journal of the ACM, 32(2), 1985 J. Turek and D. Shasha, The many faces of consensus in distributed systems, 25(6), 1992
8	2/25	Paper discussion T. Chandra, R. Griesemer, and J. Redstone, Paxos made live: an engineering perspective, In Proc. of PODC, 2007 Summary: TBD Presenter: John Hudson [slides] J. C. Corbett et al., Spanner: Google’s Globally-distributed database, In Proc. of OSDI 2012 Summary: TBD Presenter: Maciek Swiech [slides]
	2/27	Consistency and replication Replication, consistency models. [Slides pdf] Demers et al., Epidemic algorithms for replicated database maintenance, In ACM PODC, 1987. K. Petersen, M. Spreitzer, D. Terry, M. Theimer and A. Demers, Flexible Update Propagation for Weakly Consistent Replication, In Proc. of SOSP, 1997.
9	3/4	Paper discussion N. Belaramani, M. Dahlin, L. Gao, A. Nayate. A. Venkataramani, P. Yalagandula and J. Zheng, PRACTI Replication, In Proc. of NSDI, 2006 Summary: TBD Presenter: Chenjin Liang [slides] L. Gendenning, et al. Scalable Consistency in Scatter, In Proc. of SOSP, 2011. Summary: TBD Presenter: Yaonan Huang [slides]
	3/6	Fault tolerance Fault tolerance and highly available systems. [Slides pdf] Lamport, Shostak, and Pease, The Byzantine Generals Problem, ACM TOPLAS, 1982. F. B. Schneider, Implementing Fault-tolerant Services Using the State Machine Approach: A Tutorial, Computing Surveys, 1990.
10	3/11	Paper discussion M. Castro, B. Liskov, Practical Byzantine Fault Tolerance,. In Proc. of OSDI, Feb. 1999. Summary: TBD Presenter: Yixi Zhang [slides] H. Yu and A. Vahdat,The Costs and Limits of Availability of Replicated Services, In Proc. of ACM SOSP 2001. Summary: TBD Presenter: Chris Moran [slides]
	3/13	Privacy, security and censorship – paper discussion R. Dingledine, N. Mathewson and P. Syverson, Tor: The second-generation onion router, In Proc. of USENIX Security, August 1994. Summary: Georgios Tziantzioulis Presenter: Arindam Paul [slides] X. Xu, Z. M. Mao and J. A. Halderman, Internet censorship in China: Where does the filtering occur?, In Proc. of PAM, 2011. Summary: Lee Fan Presenter: Jonathan Friedman [slides]
F	3/18	Exams begin

Assignments

There are no homework assignments, only one (in two parts) team project and one take-home final exam.

That is not to say the course is easy! Students will work on teams of 2-3 on two projects over the quarter. In addition, every student will be responsible for presenting one paper from the 20 assigned for discussion (either individually, for grad students, or in teams of two for undergraduate students. Last, ever student should submit one paper summary to the blog from the same list of discussion papers; you can pick from background readings if we run out of discussion papers.

Projects

For the project component of this course you will work in teams of 2-3 students (3 is the preferred number), to implement a basic DHT (Distributed Hash Table) during the first half of the quarter, and build upon it/extend it during the second half.

The project will be done in Go, a language that was originally created within Google, but is now a fully open-source project. Go is garbage-collected and has built-in coroutines and channels, making it highly suited to building distributed systems. Its standard library is already pretty comprehensive. For example, take a look at the net and rpc packages.

Part 1

The first project - lasting six weeks - will focus on building the Kademlia DHT -- a very popular DHT first published in 2002, starting from a minimal base, following the specification linked below. You will need to keep up with the following schedule:

• Week 1: Familiarize yourselves with Go. We suggest writing a few toy programs before starting on the project proper.
• Week 4: Meet with a member of the course staff between 1/28 and 2/1 to demonstrate your progress. By this point, you should have the structure of your DHT working - new peers should be able to join your network and exchange messages with each other. The only serious work left should be implementing the rest of the protocol.
• Week 6: By 05:59 on Saturday, February 16, email a tarball containing your completed kademlia implementation to This email address is being protected from spambots. You need JavaScript enabled to view it. . Read on for a definition of "complete".

Few distributed systems operate in homogeneous environments. Your project will be evaluated based on its ability to interoperate with other kademlia implementations: if everyone implements the requirements correctly, we will be able to run a single DHT that includes clients from every group. You will receive a reference implementation at least one week before the project is due to help you ensure your code will work.

To ensure compatibility, you must strictly adhere to both the the Xlattice project's kademlia spec (linked below) and the README in the project tarball.

Part 2

During the last four weeks of the quarter, you will build on your Kademlia implementation to addresses issues, solve problems and exploit techniques from classroom discussions and readings. For example, you could enhance your DHT's ability to recover from peer churn, or you could build a consistent storage system on top of it. You should meet with the staff anytime before the fifth week (midterm) to decide on the topic of your second project.

Before starting your project, you'll need to do some reading.

• P. Maymounkov and D. Mazieres, Kademlia: A Peer-to-peer Information System Based on the XOR Metric, IPTPS 2002. The original paper.
• The XLattice Project, Kademlia: A Design Specification, Last published Feb. 2010.
• Tour of the Go language - Please work through the tour (also available as part of the SDK); you might find it worthwhile to write a simple HTTP client or other toy program before starting your work for the class.
• the provided skeleton [tar.gz]

Reading Papers and Blogging

We will be reading two or more paper per week; you should have read the assigned paper before coming to lecture. On Wednesdays, the papers will be first presented to the group by one or more students and then discussed in a round-table manner. To ensure lively discussions, you will be responsible for reading and preparing a short summary of the assigned papers before each class.

This year we will use a blog to collectively learn from everybody's insight. Everybody (other than the presenters) should read the assigned papers and blog a unique comment about the reading no later than midnight on the day before the class (to give time for everyone to read the entries before class). Note that the earlier you post, the easier it is to be unique. Entries can be parts of your summaries, e.g. main idea, the broader context for the work, a question about some aspect of the paper, an answer to someone else's question, a methodological flaw, missing related work, etc.

The blog for the site is hosted on Blogger. We will automatically add all members of the 345 Google group with @u.northwestern.edu as authors. If you do not get added, please email the TA with your Google or @u email addresses to be added as authors. Make sure to check ahead of time that you have access to the course blog.

When reading papers it is normally useful to write down a summary of about a page. Your summary should include at least:

1. Paper title and its author(s).
2. Brief one-line summary.
3. A paragraph of the most important ideas: perhaps a combination of their motivations, observations, interesting parts of the design, or clever parts of their implementation.
4. A paragraph of the largest flaws; maybe an experiment was poorly designed or the main idea had a narrow scope or applicability. Being able to assess weaknesses as well as strengths is an important skill for this course and beyond.
5. A last paragraph where you state the relevance of the ideas today, potential future research suggested by the article, etc.

You may find the following documents useful:

• How to Read a Paper by S. Keshav, ACM SIGCOMM CCR, 37(3), 2007.
• Efficient reading of papers in Science and Technology by Michael J. Hanson, 1990, revised 2000 Dylan McNamee.

Materials

Required

• Andrew S. Tanenbaum and Maarten van Steen. Distributed Systems: Principles and Paradigms. Prentice Hall, 2nd Ed., 2006 ("DSP2" from now on).
• A set of research papers I will make available (see the schedule page for a full list).

Very Useful

• S. Keshav, How to Read a Paper, ACM SIGCOMM CCR, 37(3), 2007.
• Michael J. Hanson, Efficient reading of papers in Science and Technology, 1990 (revised 2000 Dylan McNamee).
• Roy Levin and David D. Redell, An evaluation of the ninth SOSP submissions -or- How (and how not) to write a good systems paper, ACM SIGOPS Operating Systems Review 17(3):35-40 (Jul., 1983)
• George D. Gopen and Judith A. Swan,The Science of Scientific Writing,American Scientist (Nov-Dec 1990), 78: 550-558.