CS 8803: Network Measurements seminar
Fall 2003
Office: 218 GCATT
Office phone: 5-4205
Office hours: after class or by appointment
Email: dovrolis@cc.gatech.edu
Table of Contents
- Class meeting times: TR 9:35-10:55
- Classroom: 101 CCB
- Become familiar with the state-of-the-art in network measurements research
- Study the use of measurements in modeling, understanding, and improving
a network
- Learn several statistical tools for the analysis of network measurements
- Prepare and deliver a lecture on a specific network measurements topic
- Define and complete a measurements-related research project
The course will consist of two types of lectures: student presentations,
and lectures on "network statistics". Students are supposed
to study in depth one research paper prior to each class so that they
can participate in the discussion. Active participation in the
classroom is essential for this course. Additionally, a course
project with "research potential" will be carried through by
groups of 2-3 students supervised by the intructor.
- Student presentations:
Each student will give a presentation on a certain topic
of his/her interest in the area of network measurements. Students must choose
this topic by August 26. The presentation should cover an entire area of research,
and not a single paper. Typically, the area of a student presentation should
cover 5-15 research papers. A list of references used in a presentation will be provided
to the class by the presenter. Each student will have to go through his/her slides with
the instructor one week prior to the corresponding presentation.
The student presentations, as well as the related references, will be posted at
this web page. The length of each presentation should be approximately 50 minutes,
allowing for 30 minutes of discussion.
- Lectures on network statistics:
The course will also include approximately ten lectures, given by the intructor,
on a variety of statistical techniques that are often useful in the area of
network measurements. These techniques will range from basic statistical
tools (e.g., confidence intervals, tests of hypotheses) to more advanced
concepts related to long-term memory effects, heavy-tailed distributions, the
use of the wavelet transform in traffic analysis, etc. The only background
that will be assumed in these lectures is a previous course on probability
theory.
- Student projects:
Students will work in teams of 2-3 on a course project. The project topic should
be related, directly or indirectly, with the collection, analysis, or application
of network measurements. The project should have "research potential", in the
sense that with potentially some additional work after the semester is over the
project could become an original and publishable contribution. Students are encouraged
to select a project that is related to their research interests. However, the
project should not be something that they have been already working on
as part of their research. The instructor will supervise the progress of each
project during the semester, holding meetings with each student group as required.
The best projects will be submitted, if the students are interested, as research
publications in networking conferences after the semester is over.
Project milestones:
- Project description, objectives, and methodology (3-5 pages): September 11
- Progress report on intermediate results and problems (3-5 pages): October 9
- Final report, and demo if applicable (15-25 pages): December 4
Prerequisites
- CS6250 (Computer Networks) and CS7260 (Internetworking Architectures and Protocols)
or equivalent graduate-level courses.
- A course on probability theory.
The following is a list of topics that will be covered in the
course. Students are encouraged to select a presentation topic
from this list, also based on their research interests.
A reading list, made of a single paper for each topic,
will be posted at this web page as the semester progresses.
- Traffic characterization (protocol & application mix, flow sizes, packet sizes, etc)
References, Slides
- Delay and loss measurements and modeling
References, Slides
- Per-hop and end-to-end capacity estimation
References, Slides
- Available bandwidth estimation
References, Slides
- TCP-related measurements
References, Slides
- Internet traffic predictability (stationarity, "constancy", etc)
References, Slides
- Network distance estimation
References, Slides
- Internet traffic variability (long-range dependency effects)
References, Slides
- Flow measurements (counting, sampling, etc)
References, Slides
- Interdomain routing reliability measurements
References, Slides
- Interdomain routing performance measurements
References, Slides
- Measurements in overlay networks
References, Slides
- Traffic matrix estimation and applications in traffic engineering
References, Slides
- Internet topology measurements
References, Slides
- Network tomography
References, Slides
- Use of measurements in anomaly detection (flash crowds, DDOS attacks, etc)
References, Slides
- DNS measurements
References, Slides
- Web performance measurements
References, Slides
- Measurement-based admission control
References, Slides
- Measurements of peer-to-peer systems/applications
References, Slides
- Measurements of streaming/multimedia/gaming applications
References, Slides
- Tue, Aug 19, Class #1:
Introduction, course description
- Thu, Aug 21, Class #2:
Objectives, methodologies, and taxonomy of network measurements
- Tue, Aug 26, Class #3:
Network statistics (1): Overview of classical estimation theory
- Thu, Aug 28, Class #4:
Network statistics (2): Detection of heavy-tailed distributions
Reading:
Evidence for long-tailed distributions in the Internet
by Allen B. Downey (IMW 2001).
- Tue, Sep 2, Class #5:
Network statistics (3): Parameter estimation for heavy-tailed distributions
Reading:
Long-lasting transient conditions in simulations with heavy-tailed workloads
by M.Crovella and L.Lipsky
(1997 Proceedings of Winter Simulation Conference).
- Thu, Sep 4, Class #6:
Network statistics (4): From heavy tails to self-similarity
Reading:
Self-similar network traffic: An overview by K.Park and W.Willinger.
In Self-Similar Network Traffic and Performance Evaluation (ed.),
Wiley-Interscience, 2000.
- Tue, Sep 9, Class #7:
Traffic characterization
- Thu, Sep 11, Class #8:
Delay and loss measurements
- Tue, Sep 16, Class #9:
Network statistics (5): Non-parametric statistics (part I)
Reading: parts of "Nonparametric Statictical Methods", by M.Hollander and D.A.Wolf
(2nd edition), Willey.
- Thu, Sep 18, Class #10:
Capacity estimation
- Tue, Sep 23, Class #11:
Available bandwidth estimation
- Thu, Sep 25, Class #12:
Network statistics (6): Non-parametric statistics (part II)
Reading: parts of "Nonparametric Statictical Methods", by M.Hollander and D.A.Wolf
(2nd edition), Willey.
- Tue, Sep 30, Class #13:
TCP-related measurements
- Thu, Oct 2, Class #14:
Internet traffic predictability
- Tue, Oct 7, Class #15:
Network distance estimation
- Thu, Oct 9, Class #16:
Internet traffic variability
- Tue, Oct 14:
Midterm Reccess
- Thu, Oct 16, Class #17:
Flow measurements
- Tue, Oct 21, Class #18:
Interdomain routing reliability
- Thu, Oct 23, Class #19:
Interdomain routing performance
- Tue, Oct 28, Class #20:
Measurements in overlay networks
- Thu, Oct 30, Class #21:
Traffic matrix estimation
- Tue, Nov 4, Class #22:
Internet topology measurements
- Thu, Nov 6, Class #23:
Network tomography
- Tue, Nov 11, Class #24:
Network anomaly detection
- Thu, Nov 13, Class #25:
DNS measurements
- Tue, Nov 18, Class #26:
Web measurements
- Thu, Nov 20, Class #27:
Measurement-based admission control
- Tue, Nov 25, Class #28:
Peer-to-peer measurements
- Thu, Nov 27:
Holiday
- Tue, Dec 2, Class #29:
Streaming/multimedia application measurements
- Thu, Dec 4, Class #30:
TBA
- Class participation: 30%
- Topic presentation: 30%
- Project: 40%