[Dataset] pdx/metrofi (v. 2011-10-24)

top

the initial version

@MISC{pdx-metrofi-2011-10-24,
  author = {Caleb Phillips and Russell Senior},
  title = {{CRAWDAD} data set pdx/metrofi (v. 2011-10-24)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/pdx/metrofi},
  month = oct,  
  year = 2011
}
					

This is a dataset of location, signal strength, and performance data of
MetroFi, a 802.11x municipal wireless mesh network in Portland, Oregon in 2007. 
The data was collected by Caleb Phillips and Russell Senior to determine the 
coverage and performance of the network.

The data was collected during an independent analysis of MetroFi, the municipal
wireless network in Portland, Oregon.

This dataset was collected with a Linux-based sniffer using a Kismet instance 
configured to channel hop on all 11 channels. The wireless interface was an 
Atheros chipset using the madwifi driver. The 72 functional APs were located
using triangulation and then fine-tuned with a hand-held GPS device. All 
measurements are passive, with no special access to the network.

We use the standard Unix tools ttcp to test upstream throughput, ICMP ping
to test latency and loss, and wget to test downstream throughput. A small script
was used to bypass advertisement traps. We also found it necessary to use 
several watchdog scripts to check for a lost association, GPS issues, and 
stalled tests. Depending on the results, a random location test might take 
anywhere from about 60 seconds (the length of time we would wait for an 
association) to around 7 minutes. We also recorded GPS position and timestamp 
throughout the test.

The results of each test were stored on the USB storage device. At the 
conclusion of the tests we retrieved and analyzed the results.

This network is now defunct, and all the APs have been dismantled. Hence, we 
have performed no anonymization as there is no conceivable privacy/anonymity 
risk to anyone or anything.

[Traceset] pdx/metrofi/2007 (v. 2011-10-24)

top

the initial version.

@MISC{pdx-metrofi-2007-2011-10-24,
  author = {Caleb Phillips and Russell Senior},
  title = {{CRAWDAD} trace set pdx/metrofi/2007 (v. 2011-10-24)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/pdx/metrofi/2007},
  month = oct,  
  year = 2011
}
					

This is a traceset of location, signal strength, and performance data of
MetroFi, a 802.11x municipal wireless mesh network in Portland, Oregon in 2007.
The data was collected by Caleb Phillips and Russell Senior to determine the 
coverage and performance of the network.

Because our tests were carried out without any access to the network infras-
tructure, our first task was to locate the access points in the POC area. To 
this end, we drove to every publicly accessible street, collecting signal 
strength measurements using a battery-powered embedded computer with an 
external 7 dBi omnidirectional antenna and a GPS device. We used this data to 
triangulate the position of the APs. Not surprisingly, as other researchers 
have shown that signal strength is poorly correlated with distance, we were 
unable to reach a satisfactory level of precision. To obtain the desired 
precision, we used triangulation information to locate each AP and then took a 
reading with a hand-held GPS device directly under the AP.

From the list of 72 MetroFi access points that we considered to be in the POC 
network, we constructed a bounding box in latitude and longitude extending 1000
feet beyond the extremities of the access point locations. Because we expected 
that many locations in the bounding box would fall outside of the POC areas, 
and because we were not certain how many locations we would be able to measure,
we computed an excessive sample of 1001 locations using a random number 
generator such that each location in the bounding box had an equal probability 
of being chosen. Locations not within 1000 feet of an access point were 
immediately excluded. Each remaining location was plotted against orthoimagery 
using Google Maps. If the location fell in the Willamette River, was inside a 
building, or was not practically reachable, it was also excluded. Ultimately, 
the first 250 locations in our sample of 1001 were either excluded on the basis
of the criteria above or were visited and measured. We chose to stop at 250 
points after finding that this well bypassed our needs in terms of statistical 
power, both in the POC and overall.

We use the standard Unix tools ttcp to test upstream throughput, ICMP ping
to test latency and loss, and wget to test downstream throughput. A small script
was used to bypass advertisement traps. We also found it necessary to use 
several watchdog scripts to check for a lost association, GPS issues, and 
stalled tests (for example, ttcp has a tendency take a very long time on 
unstable connections). Depending on the results, a random location test might 
take anywhere from about 60 seconds (the length of time we would wait for an 
association) to around 7 minutes. In addition to these steps, we also recorded 
GPS position and timestamp throughout the test.

We have performed no anonymization as there is no conceivable privacy/anonymity 
risk to anyone or anything.

[Trace] pdx/metrofi/2007/coverage (v. 2011-10-24)

top

the initial version

@MISC{pdx-metrofi-2007-coverage-2011-10-24,
  author = {Caleb Phillips and Russell Senior},
  title = {{CRAWDAD} trace pdx/metrofi/2007/coverage (v. 2011-10-24)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/pdx/metrofi/2007/coverage},
  month = oct,  
  year = 2011
}
					

This is a trace of location, signal strength, and performance data of MetroFi, 
a 802.11x municipal wireless mesh network in Portland, Oregon in 2007.

This dataset was collected with a Linux-based sniffer using a Kismet instance 
configured to channel hop on all 11 channels. The wireless interface was an 
Atheros chipset using the madwifi driver. The 72 functional APs were located
using triangulation and then fine-tuned with a hand-held GPS device. All 
measurements are passive, with no special access to the network. 

We use the standard Unix tools ttcp to test upstream throughput, ICMP ping
to test latency and loss, and wget to test downstream throughput. A small script
was used to bypass advertisement traps. We also found it necessary to use 
several watchdog scripts to check for a lost association, GPS issues, and 
stalled tests. Depending on the results, a random location test might take 
anywhere from about 60 seconds (the length of time we would wait for an 
association) to around 7 minutes. We also recorded GPS position and timestamp 
throughout the test.

Besides the README file, there are three pieces of data:

1) aps.txt: Locations and MAC addresses of APs

2) stumble_filtered.txt: Received signal strength indicator (RSSI) observations
   of the APs from the exhaustive war-driving campaign in the network's service
   area. To turn RSSI into something like Received Signal Strength (RSS) in 
   dBm, simply subtract 95.

3) random.csv: Performance results at a random sample of points within the 
   coverage area. The fields are listed on the first line.

We have performed no anonymization as there is no conceivable privacy/anonymity 
risk to anyone or anything.

[Author] Caleb Phillips

top

[Author] Russell Senior

top

[Paper] phillips-robust

top