High-Performance Networking

Course Description

EECS 881
3 credit hours

Comprehensive coverage of the discipline of high-bandwidth low-latency networks and communication, including high bandwidth-×-delay products, with and emphasis on principles, architecture, protocols, and system design. Topics include high-performance network architecture, control, and signalling; high-speed wired, optical, and wireless links; fast packet, IP, and optical switching; IP lookup, classification, and scheduling; network processors, end system design and protocol optimization, network interfaces; storage networks; end-to-end protocols, mechanisms, and optimizations; and high-bandwidth low-latency applications. Principles will be illustrated with many leading-edge and emerging protocols and architectures.

Prerequisites

EECS 780, 563, or 663.

Time and Location

EECS 881 meets one evening a week for three hours on the Edwards Campus in the western Kansas City suburb of Overland Park. A lab section session is scheduled on an additional evening and will meet as needed for introduction and guidance in completion of the ONL exercises, as well as for class make-up and exam reviews. See the individual course offering pages for detailed time and room information.

For Kansas City residents, this is 2.4 mi. south from the Quivera Road exit of the southwestern portion of the I-435 loop to 127th St. For Lawrence residents this is approximately 30 miles / 50 km east of the Lawrence Campus, a 40 minute drive mostly along the K-10 freeway. A parking permit is not needed on the Edwards Campus. The K-10 Connector bus is a service used by many students between the Lawrence and Edwards campuses.

Course Offerings

Detailed information about individual offerings of this course will be located on the following pages, including schedule and homework assignments.

• fall 2010: KU EECS 881 (planned for Edwards Campus)
• fall 2008: KU EECS 800 (Edwards Campus)
• fall 2006: KU EECS 800 (Edwards Campus)
• spring 2004: Universität der Bundeswehr München Germany graduate diploma-level intensive course
• 2002–2006: IEEE HotI conference tutorial
• 2003: IEEE MILCOM conference tutorial
• 2002: IEEE ICN/ICWHLN conference tutorial

Generic course information and the latest version of the lectures are located this page below.

Lectures and Readings

Note: this table is currently under reorganisation

Reading assignments: S = Sterbenz & Touch, C = Comer, V = Varghese, W = Welzl, K = Kurose & Ross

Textbooks

Required Textbooks

Important Notice: There are changes in the required and optional textbooks this year.

These textbooks and the corresponding required readings in the table are essential for success in this course. Students are responsible for knowing all of this material regardless of whether or not explicitly covered in lectures.

James P.G. Sterbenz and Joseph D. Touch,
High-Speed Networking: A Systematic Approach to High-Bandwidth Low-Latency Communication,
John Wiley, New York, 2001.
This book is also used for EECS 780

Douglas E. Comer,
Network Systems Design using Network Processors, Intel IXP 2xxx Version,
Pearson Prentice-Hall, 2006.

Optional Textbooks

These textbooks will contain optional reading assignments that should contribute significantly to your understanding of the material. In an effort to keep student costs down, they will be placed on library reserve, however students may find it more convenient to purchase their own copies.

George Varghese,
Network Algorithmics,
Morgan-Kaufmann Elsevier, 2005.

Michael Welzl,
Network Congestion Control,
John Wiley, Chichester UK, 2005.

Supplementary Textbooks

[RS2002]
Rajiv Ramaswami and Kumar N. Sivarajan,
Optical Networks: A Practical Perspective,
second edition, Morgan Kaufmann, 2002.

[J2003]
Thomas C. Jepsen,
Distributed Storage Networks: Architecture, Protocols and Management,
Wiley, 2003.

[TMK2006] Franco Travostino, Joe Mambretti, Gigi Karmous-Edwards, ed.,
Grid Networks: Enabling Grids with Advanced Communication Technology,
Wiley, 2006.

Background Textbook

This book is one of the textbooks used in EECS 780 and is an excellent source of background material for this course.

James F. Kurose and Keith F. Ross,
Computer Networking, fourth edition,
Pearson Addison Wesley, 2008.

Additional Readings

The readings in this section are listed with the lecture to which they correspond, before which all students must read the papers. Students may request a particular paper or topic in advance by email.

The presenting student should construct a presentation that takes 20 minutes to deliver without questions, with a maximum of 10 content foils (excluding title, outline, and reference foils). If you are inexperienced, you must practise to a mirror or friend until you've got the timing right. Significant points will be deducted for presentations that are too long. You should read the presentation guidelines before you create your presentation, and use the template as a basis for your presentation. You must not use font sizes smaller than this template, and are strongly urged to turn off auto shrink to fit to avoid font size problems. If you are an experienced presenter with a PowerPoint style you are comfortable using, or if you wish to use other programs to create your presentation, you may do so with prior approval only after I have seen a sample, and this must be done well in advance of your scheduled presentation. Otherwise, you must use the provided template. Presentations must be emailed the day before the scheduled delivery (by 23:59) in PDF (recommended) and PowerPoint source (unless we have agreed on another presentation tool) with a Subject: line beginning EECS881 - reading presentation: Once a presentation has been scheduled, it is not possible to reschedule for any reason other than emergency, to avoid disrupting the class schedule.

A one-half-hour time slot will be devoted to each presentation to allow for questions from other class members, which contributes to the participation grade.

[BHP2006]
Ashwin R. Bharambe, Cormac Herley, and Venkata N. Padmanabhan,
“Analyzing and Improving a BitTorrent Network’s Performance Mechanism”
Proceedings of IEEE INFOCOM 2006,
2006, pp. 1–12

[FB+2005]
W. Feng, P. Balaji, C. Baron, L. N. Bhuyan, D. K. Panda,
“Performance Characterization of a 10-Gigabit Ethernet TOE”
13th IEEE Symposium on High Performance Interconnects (HOTI'05), 2005,
2005, pp. 58–63

[HOT1997]
John Heidemann, Katia Obraczka, and Joe Touch,
“Modeling the Performance of HTTP Over Several Transport Protocols”
IEEE/ACM Transactions on Networking (TON),
vol.5 iss.5, 1997, pp. 616–630

[J2006]
“Juniper Networks Router Overview”
JUNOS Baseline Network Operations Guide, Juniper Networks,
2006, pp. 3–15

[J2008]
Multicast Architectures in Crossbar-Based Routers: Replication Choices and their Impact on Real-World HDTV, Juniper Networks,
white paper 200281, 2008

[PK1998]
Venkata N. Padmanabhan and Randy H. Katz,
“TCP Fast Start: A Technique For Speeding Up Web Transfers”
Proc. IEEE Globecom ‘98 Internet Mini-Conference,
1998, pp. 41–46

[QY1999]
Chunming Qiao and Myungsik Yoo,
“Optical Burst Switching (OBS) – A New Paradigm for an Optical Internet”
Journal of High Speed Networks, IOS Press,
vol.8 iss.1, March 1999, pp. 69–84

[SBD2001]
Miguel Á. Ruiz-Sánchez, Ernst W. Biersack, and Walid Dabbous,
“Survey and Taxonomy of IP Address Lookup Algorithms”
IEEE Network,
vol.15 iss.2, March/April 2001, pp. 8–23

[SL2005]
Haoyu Song and John W. Lockwood,
“Efficient Packet Classification for Network Intrusion Detection using FPGA”
ACM International Symposium on Field-Programmable Gate Arrays (FPGA'05),
February 2005, pp. 238–245

[T2005]
David E. Taylor,
“Survey & Taxonomy of Packet Classification Techniques”
ACM Computing Surveys,
vol.37 iss.3, September 2005, pp. 238–275,
preprint Washington University Computer Science and Engineering technical report WUCSE-2004-24

[X2005]
Yang Xiao,
“IEEE 802.11n: Enhancements for Higher Throughput in Wireless LANs”
IEEE Wireless Communications, IOS Press,
vol.2 iss.6, December 2005, pp. 82–91

Grading

Grading will be on a modified curve in which students are grouped (generally by modes in the distribution). Exams and homework will receive numerical scores; the term paper will receive a letter grade (with + and – discriminators) that will be converted to a numeric value for determining final weighted average. Final grades at KU do not have the + and – modifiers. Employer reimbursement and immigration status cannot be a consideration in the final grade.

If you are having difficulty in the class I strongly recommended you discuss this early and not wait until exam time. Students are responsible for understanding course drop policies and deadlines.

Assignments and Exams

ONL Laboratory Exercises and Homework Problems

Homework problems will occasionally be assigned.

ONL laboratory exercises will permit remote experimentation on real high-speed switch hardware and network processors.

Specific assignments and dates are located in the course offering page for a given semester. Assignments will not be graded nor receive credit unless they follow the submission requirements.

Exams

Exams will be closed book and take approximately 1/2 of a class period. The exam information page contains detailed information on the requirements, structure, and grading of examinations for this course. You must also read the academic integrity page before taking an exam.

While you are responsible for all lecture and required readings, the following list outlines some of the most important topics likely to be covered on the exams for this course.

Exam 1: Architecture and Lower Layers

Architecture and lower layers

• understanding of the components of delay and bandwidth-×-delay product and the constituants of delay in networks
• understanding the end-to-end arguments, and the guidance it gives to functional placement in the network
• understanding of the systemic optimisation principle, its variants and corollaries, and the guidance it gives to high-performance design
• understanding the various signalling paradigms and their relative strengths and weaknesses relative to delay and throughput
• physical media and high-speed links
• evolution of 802.3 and 802.11 and modifications made for higher data rates

Exam 2

Middle layers

• scalabiity of SONET/SDH and OTN
• evolution of 802.16 and 802.15
• evolution of DSL and HFC
• network processor functions and parallelsim vs. pipelining
• fast packet switching and label swapping
• packet size and structure with respect to performance
• switch fabric architecture (crossbar, MIN, and Clos) and blocking characteristics
• contention, input, output, and virtual output queuing
• native mulitcast switch fabrics
• fast datagram switches and LPM IP address lookup techniques
• packet classification
• packet scheduling alternatives
• end system bottlenecks and software optimisations (e.g. threads, kernel-crossing avoidance, memory mapping, protocol bypass, ILP
• host interconnects and relationship to network interfaces
• network interface architecture, protocol offloading, TOEs

Exam 3

Upper layers

• functions and mechanisms of the transport layer and how they impact performance, including:
  • open vs. closed loop error and transmission rate control
  • impact of high speed and long latency on state management
• various error control mechanisms and their relative performance (including side-by-side comparison of stop-and-wait, go-back-n, and selective repeat), as well as key optimisations such as fast retransmit
• congestion control in general (including throughput and delay curves), and as implemented by TCP in particular
• the three application classes (information access, telepresence, distributed computing and storage) and their characteristics
• utility curves for delay-sensitivity and the difference between a best-effort application and a best-effort network service
• the meaning and implication of individual vs. aggregate bandwidth measures for an application
• HTTP and optimisations to reduce latency
• the performance implications of the various generations of P2P file sharing
• mirroring, caching, and CDNs to reduce latency and bandwidth
• in general terms: networked storage and distributed computing
• file swarming, and how it differs from previous P2P file sharing

Final Exam

Comprehensive

• material from any part of the course
• principles on high-performance design from Chapter 2, in general and as applied in specific contexts we've covered
• questions that synthesise multiple layers or parts of the course

Navigation: Up: courses – Previous: EECS 780 – Next: EECS 882 – Top: James P.G. Sterbenz

Last updated 24 November 2008 – Valid XHTML 1.1 – Lynx inspected – W3C AAA Conformance
©2006–2008 James P.G. Sterbenz <jpgs@eecs.ku.edu>