Note that this information is old and does not reflect the current class. All project related
postings for the current class is available on Blackboard.
CSE 570 Course Project
Ground
Rules
- Two students per project. Three students team is
acceptable; but more comprehensive work will be expected (naturally).
Going solo is also acceptable; but not recommended. It usually means a
lot of work for the individual to complete a reasonable project.
- Your goal should be developing an idea to solve a
problem, and evaluating your idea. Most common method of evaluation is
via simulation. But sometimes analytical methods or experiments are
appropriate. Evaluating existing techniques (e.g., comparative
evaluation of existing techniques to solve a specific problem) can also
make good projects. You should read of a lot of papers related to your
project topic and attempt to discover better/more comprehensive
solutions/approaches/studies than may exist in literature.
- The project report should resemble a paper.
There is no specific page limit or format requirements. The best
projects are likely to result in actual conference/workshop papers,
possibly with some additional follow-up work later. This should be your
goal.
- It is expected that each project will be presented
in a 15 min presentation in class. We will come up with a presentation
schedule later. If results are not available at the time of
presentation, the presentation should focus on the background, concept,
design, related published work, etc. Here, your goal will be to educate
others about your work. If there are too many projects than we have
time for, we will either pick only certain projects to be presented, or
use a poster session.
- The project grade will be based on the
difficulty/depth of the problem you are trying to solve, your
accomplishments/progress on solving the problem, and your
report/presentation. This means that an easy
problem will require a very comprehensive work, while less complete
work will be acceptable for a challenging problem. If you are able to
come up with a new and promising idea, even initial work will be
appreciated. Basically, you are do not
easily win by choosing an easy problem.
- To make sure that you
are actually working on the project at a steady pace, we will have
checkpoints. The checkpoints are not mandatory, and are only meant to
give you feedback. But if you receive poor feedback on checkpoints, you
should take it seriously. Though checkpoints are not mandatory,
experience has shown that those who do not submit checkpoints, receive
poor grades in the final submission. Checkpoint reports also help you
with the final report as you can reuse much of the text you have
already written. Checkpoints should be emailed with subject header "Checkpoint n: CSE570," where n is 1 or 2.
- Checkpoint 1: Due XXX. Submit a 0.5-1 page document describing
describing the broad idea of your project with appropriate references.
- Checkpoint 2: Due XXX. Submit a 2-2.5 page document describing a
detailed set of goals and an initial design of your project. This
should contain enough details to convince us that you have a good set
of plans for a successful project, and you have a focus. For example,
it should contain a high-level description of the
techniques/algorithms, choice of evaluation mechanisms/platforms etc,
probably some initial experiments/results.
Project Ideas
We have ideas here that you can
use as starting points. Note that the ideas
as stated may be broad-based or vague or incomplete or too complex. It
is your
responsibility to pick an idea or part of an idea and make it somewhat
concrete.
It is acceptable to have more than one team working on the same broad
idea. We
are sure that you will all have different approaches.
You are also encouraged to
work on your own idea that is not in the list.
For most ideas we have one or more
references to get you started. But many of them may have more
substantial
literature.
Layer 3 and Up
- Geographic Multicast: How would you use geographic routing to perform
multicast? How much is the performance advantage relative to
traditional methods? [Some ideas are presented in
this Secon 06 paper or in this
Position Based Multicast Routing paper.]
- Efficient
Content Caching/Access in Vehicular Networks: Imagine
content access from cars. Most of the content is expected to be similar
for the cars that are in the neighborhood. For example, the content
could be maps, weather, location-based ads etc. Using cellular networks
for content access (using a cellular data link in the car) could be
expensive. On the other hand, it is possible to distribute such content
using a combination of caching and ad hoc network routing. The idea is
basically that only some cars download content using cellular links.
Content is cached. Other cars access such content using 802.11-based ad
hoc networks. All cars do carry both 802.11 and cellular data
interfaces. However, to reduce cost, the cellular links are used only
sparingly. Design and evaluate an efficient (i.e., low latency) scheme
for such networks that optimize the use of cellular links. Your
position should be that of a service provider that specializes in
providing low cost network connectivity to cars and in turn contracts
cellular links from a cellular provider and pays a per bit usage cost
to this provider. [Reference: Bin Tang's
work in WINGS Lab web page.]
Layer 2 and Below
- Bad fish problem: In traditional CSMA protocol (such as in 802.11) all
nodes get an equal chance to access the radio medium. Thus, nodes with
slower bit rate may get an unfair share of the bandwidth as they take a
longer time to transmit a packet. These nodes could be those with
already poor links (e.g., too far from the AP.) Characterize this
problem and suggest solutions. Solutions can range from fixing the
802.11 protocol itself (ideal) to deploying practical solutions.
[Reference:
This Infocom03 paper.]
- Misbehavior in the MAC
layer:
Characterize the misbehaviors that are possible in the 802.11 MAC layer
that can either give the misbehaving node unfair (as defined by the
802.11 protocol) share of the network bandwidth or can launch a denial
of service attack in an extreme case. Suggest and evaluate approaches
to detect such misbehaviors. [Reference: Nitin Vaidya's work in UIUC and Jean-Pierre
Hubaux's work in EPFL (Domino protocol). Utpal Paul's work in WINGS Lab
web page.]
- MAC layer (reliable)
multicast: How
would you do multicast in the MAC layer? Using simply broadcast may not
be a good idea as typical broadcasts are not reliable, as there is no
acknowledgement. This is important as broadcasts may suffer from the
hidden terminal problem. What is the performance impact of the
reliability mechanism? [Reference: Mario Gerla's work in UCLA. Shweta Jain's
work on WINGS lab web page. The
DirCast work in MSR ]
- Power control/Carrier
sense threshold/Rate control in CSMA protocol: One or more of these three
parameters can be controlled (statically/dynamically) to optimize
performance. You can use the 802.11 protocol or a simpler variation and
study the impact of controlling one or more of these on performance,
and may be devise strategies or protocols to control these. [Reference: Nitin Vaidya and his
students' work (UIUC). Also look at Jennifer Hou (UIUC) and her
students' work].
- Channelization for High
Speed Wireless Networks: If the data
rate of the wireless network increases, packet transmission times
become shorter. However, as carrier sense and propagation time
overheads do not change with data rate, these overheads look relatively
larger, thus reducing the efficiency of the network. If you consider
your homework 2, the net effect of this would be slots remain the same,
but packet size (in slots) is small. The options to tackle this problem
is to increase packet size (coalescing packets - used in 802.11n) or
using multichannel. In multichannel, the available bandwidth is split
into multiple smaller channels and transmissions concurrently happen in
all these channels. Smaller channels mean that packet size (in slots)
are now larger. You can understand this problem via simulations
and address via channelization. [Reference: This
paper in Sigcomm10 describes and addresses the core problem
using multiple channels.]
- Localize Roadside WiFi AP from Wardriving
Data: We have a large amount of sniffed
WiFi traces collected in the course of
this paper. Note that much of the trace corresponds to repeated
drives on the same roads. Localize the WiFi APs seen in this trace. You
can try a few methods and evaluate the difference between them. For
example, you can localize an AP at the centriod of all the locations
you have heard beacons from this AP. Based on the localization lecture,
you may be able to do better by incorporating the signal strength and
may be a propagation model. (Contact Pralhad Deshpande for the trace.
You will need to develop a pretty good idea of 802.11 frame types and
formats to make sense of the trace.)
The following are primarily experimental/systems
projects. These usually work better if you have some background on
related systems or have a hacker attitude. I can help with hardware.
Because of limitation of available hardware, I will have to limit such
groups to only a few.
- SINR modeling on
sensor motes: Experimentally
develop a model for bit/packer error rate vs. SINR on the motes
platform. You need to learn how to work with motes and TinyOS. [References: Similar work has been
done by Bhaskar Krishnamachari (USC) in SECON05 and Sensys06, but using
a different hardware than what we have. Ritesh Maheshwari in WINGS lab
will help].
- WiFi Traffic Analysis: Collect WiFi traffic via snooping
in a busy environment. Analyze the data. Comment on performance. Are
there performance issues? Any interesting observations? Multiple
concurrent snoopers will make the work more interesting. [Reference: Elizabeth
Belding-Royer's work (UC-Santa Barbara) in IMC 05 and IMC07. David
Kotz's (Dartmouth) work in MobiCom02 and Mobicom04. Also, Ratul
Mahajan's (Microsoft Research) work in Sigcomm06].