Research Plan

Alok Shriram is a PhD student studying with Jasleen Kaur at the University of North Carolina - Chapel Hill. During summer 2004, Alok is performing e2e bandwidth estimation research at CAIDA. His research plan is below.

| Project Home Proposal Goals Presentations People Test_Lab Reports Related Work |

Research Sponsored by:

Alok Shriram's Summer 2004 Research Plan

Tools to test:

Pathload
Pathchirp
Spruce
Abing
Iperf

Tools not to test any more (based on their prior substandard performance):

ABw
bprobe
cprobe
clink
pathchar
pchar
pipechar
tracerate

Laboratory measurements

Perform three sets of experiments using different types of cross-traffic. Test each tool in both directions.

SmartBits - artificially generates traffic, maintaines a predefined load level
- Run tests increasing the load in 10% steps from 10% up to 90%.
- Each tool should run continuously for a given amount of time (currently, 5 min).
- Plots to make:
  - for each tool: the average & standard deviation of obtained values vs. actual available bandwidth, both directions on one graph (5 graphs combined in one panel)
  - tool run time vs. actual available bandwidth (all tools in one graph)
  - "consumed bandwidth"
  - # of bytes sent by tool divided by run time vs. actual available bandwidth (all tools in one graph)
  - IAT distribution of SmartBits traffic
  - packet size distribution of SmartBits traffic
tcpreplay - plays back a real traffic sample collected elsewhere.
- Use the same sample for all tools.
- Estimate the load level using CoralReef/NeTraMet.
- Each tool should run continuously for a given amount of time (currently, 5 min).
- Plots to make:
  - actual available bandwidth (per 1 sec) and each individual measurement for each tool vs. time (possibly, all tools in one graph, but if it will be too crowded, then change somehow)
  - IAT distribution of tcp-replay traffic - packet size distribution of tcp-replay traffic
HTTP request generator - simulates HTTP traffic from a varying number of browsers.
- Use the same set of conditions for all tools.
- Estimate the load level using CoralReef/NeTraMet.
- Each tool should run continuously for a given amount of time (currently, 5 min).
- Plots to make:
  - actual available bandwidth (per 1 sec) and each individual measurement for each tool (possibly, all tools in one graph, but if it will be too crowded, then change somehow)
  - IAT distribution of simulated HTTP traffic
  - packet size distribution of simulated HTTP traffic

Real-Path Measurements

We will run the best performing (in the lab) tools on the ESnet network.

Obtain accounts on two hosts in the ESnet network, at SLAC and at ORNL.
Install tools.
Do a single "pre-test" run of each tool.
Run measurements.
- Possible schedule: run each tool 10 times every hour for 24 hours.
- Other schedules TBD.
How can we get an independent measure of real ESnet traffic in order to calibrate/verify our results?