Project for Capacity Planning
Graduate students only in Capacity Planning are required to
complete a project. On this page are some topics to "get you started".
I am open to new project ideas.
The project is expected to be something significant. The final
deliverable must be in standard conference paper format. An example of
a project from Spring 2001 is rogers.pdf. This
paper will appear in the IEEE LCN
conference in November 2001. This conference has a 50% acceptance rate.
A template for a standard IEEE format paper is
sample.pdf and in "source code form"...
sample.doc. You must exactly follow this format.
The first deliverable is a one page proposal (due on 09/13) containing:
- A description of the problem
- A description of how the problem will be solved
- At least five references relevant to the proposed project
Possible topics include...
- It is difficult to make one-way delay measurements in the Internet. There
are a lot of two-way (i.e., round-trip time) delay studies, but few of
one-way delays. For this project you are to develop a simple tool that
sends time-stamped packets via UDP to a "catcher" program. The catcher program
writes the send and receive times of each received packet to a file. Once the
tool is written, you are to deploy the tool between two distant sites and
collect several days of measurements. You are finally to investigate
correlation between one-way and two-way delays.
- Video traffic is expected to continue to increase in volume on packet
networks. What does "video cam" traffic look like? What are its
characteristics for packet size and interarrival time? For this project you are
to install at least two different video cams and trace the generated network
traffic. Then you are to characterize the collected traffic traces.
- I have a trace of 5 million packets collected from the USF backbone. From
this trace I built an empirical distribution for generating packet lengths.
This first order models assumes that there is no correlation between packet
lengths. Is this a correct assumption? You are to explore packet length
characteristics in this trace file, and other traces. You are to come-up with
a simple packet length model that is better than the first order empirical
distribution model.
- The ping tool measures delay. Can you come-up with a tool that measures
(or estimates) bandwidth? Here is a simple (and well known) method. Send
two back to back packets as a "ping". Then, measure the gap between the
packets on their return. Some thinking should have you see that this gap
will be caused by lower bandwidth links that the originating sending link.
From this gap you can estimate the bandwidth of the slowest link in the
path.
- The ping tool sends out pings periodically (deterministically). However,
a Poisson view (PASTA = Poisson Arrivals See Time Averages) may give a more
accurate (or "real") view of RTT. Modify a ping program to have exponentially
distributed interarrival times and then compare your and existing ping for
a number of paths. Is there are difference in mean RTT? If there is, how can
we test which answer is "correct"?
Last updated by
Ken Christensen on AUGUST 23, 2001