FastMapping data set

This page describes the data set used in the paper:

Measuring and Characterizing End-to-End Route Dynamics in the Presence of Load Balancing
I. Cunha, R. Teixeira, and C. Diot
Passive and Active Measurement Conference
Atlanta, GA, March 2011.
[pdf]

This data set contains traceroute-style measurements from monitors to 1,000 destinations using FastMapping, a probing method that exploits load balancing characteristics to reduce the number of probes needed to measure accurate route dynamics.

We used 70 PlanetLab nodes as monitors during five weeks starting September 1st, 2010. Click here for the list of PlanetLab nodes we used. Each monitor selects 1,000 destinations at random from a list of 34,820 randomly chosen reachable destinations. Monitors use ICMP probes and probe as fast as they can, which results in an average measurement round duration of 4.4 minutes. For a more detailed description of FastMapping, please see the original paper (PDF).

The data is organized first by monitor, then by measurement round. A measurement round comprises the measurement of all monitored paths. The result of each measurement round is stored in a file named <timestamp>.gz, where the timestamp marks the time when the measurement round started. The result of each measurement round is stored in binary form. We provide example code that reads the binaries and provides an interface to the data.

Example

monlist.txt
The list of 70 PlanetLab nodes used in the data set.

dstlist.txt
The list of 34,820 destinations from which monitors choose 1000 destinations to probe.

dlib.py
Python code with an API to load the binary data set in memory and access it. See an example of usage below. The code has inlined Python documentation.

chronos-day1.tar
This file contains data for the first day of measurements from the monitor at chronos.disy.inf.uni-konstanz.de. This file contains several <timestamp>.gz files described above. To load the 1283506986.gz file and print its contents, it is enough to use the commands below in a Python script:

import gzip
from dlib import Snapshot
fd = gzip.open('1283506986.gz')
s = Snapshot.read(fd)
print s

Dataset download

The whole data set (with the five weeks of data from the 70 monitors) is 47GB. As we have limited bandwidth, we currently request interested researchers to: first, use the example trace and scripts given above to evaluate the utility of the traces for your purposes; then, send an email to Italo Cunha to get access to the complete traces.

We are currently working towards a more convenient solution. If you have enough bandwidth to mirror the data, we would be very happy to upload it and link it from this page.

Contact

Italo Cunha <lastname dot remove dot this at dcc dot ufmg dot br>