[Dataset] rutgers/ap_density (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} data set rutgers/ap_density (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density},
  month = aug,  
  year = 2007
}
					

This dataset includes realistic client arrival patterns and realistic application 
workloads that are used to evaluate effects of high-density deployments of 
interfering access points to IEEE 802.11 WLAN performance. Both wired and 
wireless frame dump traces are provided from the experiments in ORBIT wireless 
research testbed involving four access points and seventy-five stations.

Experiments are conducted to measure effects of having multiple 
interfering APs that serve many users with realistic traffic workloads.

The experiments were performed on the 400-node grid of ORBIT, wireless research testbed. 
Placement of nodes follow a high density grid-type deployment in a 20 meters by 20 meters 
isolated area with approximately 1 meter inter-node distances.

ORBIT provides access to small form factor PCs with 1GHz Via C3 CPU, 512 MB RAM, 
20 GB hard disk, three Gbps ethernet ports, and two Atheros AR5002X Mini PCI 802.11a/b/g 
(AR5212 MAC Baseband Processor + AR5112 Dual-band Radio-on-a-Chip) WLAN adapters. 
We have used GNU/Linux operating system with 2.6.18 series kernel, and Madwifi driver 
svn revision 2118. For access points, we have used Madwifi driver's AP mode 
with transparent bridging to that particular ORBIT node's Gbps Ethernet interface. 
All application services (WWW, and D-ITG -- VoIP and Exponential traffic) are also 
hosted on ORBIT nodes that are connected to the same switched Gbps data backbone.

[Traceset] rutgers/ap_density/workload (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-workload-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace set rutgers/ap_density/workload (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/workload},
  month = aug,  
  year = 2007
}
					

This traceset includes realistic client arrival patterns and realistic application 
workloads that are used to evaluate effects of high-density deployments of 
interfering access points to IEEE 802.11 WLAN performance.

The experiments are conducted to measure effects of having multiple interfering APs 
that serve many users with realistic traffic workloads as follows:

- Clients follow an arrival pattern to associate with the AP(s). The station arrival pattern
was created similar to the real users returning from lunch at the 64th IETF conference. 
(The trace rutgers/ap_density/workload/sta_arrivals contains the arrival pattern.)

- As clients arrive, they select an application profile (WWW, exponential traffic, or VoIP) 
randomly to match the percentage of traffic distribution that is observed in a typical 
office setting. After selecting their profiles, clients start the application session and 
generate traffic until the end of the experiment. 
(The trace rutgers/ap_density/workload/realisticTraffic contains the traffic workload.)

[Traceset] rutgers/ap_density/wired (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-wired-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace set rutgers/ap_density/wired (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/wired},
  month = aug,  
  year = 2007
}
					

This traceset includes wired frame dump traces collected from the experiments 
in ORBIT wireless research testbed involving four access points and seventy-five 
stations.

The experiments are conducted to measure effects of having multiple 
interfering APs that serve many users with realistic traffic workloads. 
The AP and clients use IEEE 802.11a protocol on channel 52 which is 
confirmed to be unoccupied throughout the experiment. Clients follow 
an arrival pattern (see rutgers/ap_density/workload/ap_arrivals) to associate 
with the AP(s). 
If more than one AP is available, default AP selection algorithm 
on driver is used (highest RSSI AP gets picked). As clients arrive, 
they select an application profile (WWW, exponential traffic, or VoIP) 
randomly to match the percentage of traffic distribution that is observed 
in a typical office setting. After selecting their profiles, clients start 
the application session and generate traffic until the end of the experiment. 
Cumulative system throughput (as measured flowing between APs and application 
server) determined system performance in hauling traffic back and forth between 
wired and wireless networks. Multiple runs of the same experiment resulted 
in similar behaviors, thus only the traces from one instance of 
each experiment are made available to public.

[Traceset] rutgers/ap_density/wireless (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-wireless-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace set rutgers/ap_density/wireless (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/wireless},
  month = aug,  
  year = 2007
}
					

This traceset includes wireless frame dump traces collected from the experiments 
in ORBIT wireless research testbed involving four access points and seventy-five 
stations.

The experiments are conducted to measure effects of having multiple 
interfering APs that serve many users with realistic traffic workloads. 
The AP and clients use IEEE 802.11a protocol on channel 52 which is 
confirmed to be unoccupied throughout the experiment. Clients follow 
an arrival pattern (see rutgers/ap_density/workload/ap_arrivals) to associate 
with the AP(s). 
If more than one AP is available, default AP selection algorithm 
on driver is used (highest RSSI AP gets picked). As clients arrive, 
they select an application profile (WWW, exponential traffic, or VoIP) 
randomly to match the percentage of traffic distribution that is observed 
in a typical office setting. After selecting their profiles, clients start 
the application session and generate traffic until the end of the experiment. 
Cumulative system throughput (as measured flowing between APs and application 
server) determined system performance in hauling traffic back and fort between 
wired and wireless networks. Multiple runs of the same experiment resulted 
in similar behaviors, thus only the traces from one instance of 
each experiment are made available to public.

[Trace] rutgers/ap_density/workload/sta_arrivals (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-workload-sta_arrivals-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/workload/sta_arrivals (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/workload/sta_arrivals},
  month = aug,  
  year = 2007
}
					

This trace contains station inter-arrival times in seconds at each line 
for a total of 75 stations.

This file contains station inter-arrival times in seconds at each line 
for a total of 75 stations. A control loop like:

for (i=1;i<75;i++)
{
create_station(i);
sleep( read_line(i, sta_arrivals.txt) );
} 

will create station arrival patterns similar to the real users returning 
from lunch at the 64th IETF conference. This arrival pattern is extracted 
from the UCSB 64nd IETF conference traces. In processing those traces, 
a station is assumed to be inactive if no IP traffic (both directions) is 
detected for more than five minutes for that particular node. First 75 inactive 
nodes that start communication after the lunch hour are captured in the trace file.

Plain Text (UNIX CR/LF)

[Trace] rutgers/ap_density/workload/realisticTraffic (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-workload-realisticTraffic-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/workload/realisticTraffic (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/workload/realisticTraffic},
  month = aug,  
  year = 2007
}
					

This trace contains a synthetic office work-load based on sniffer measurements 
obtained in our academic office environment for a single access point serving 
about 50 students and faculty.

Since inter-cell interference is affected by end-user workloads, 
we constructed a synthetic office work-load in addition to bulk TCP-only workloads 
we used in PER experiments. The office workload is based on sniffer measurements 
obtained in our academic office environment for a single access point serving 
about 50 students and faculty. These measurements indicate that 97% of packets 
use the TCP protocol and about 75% of traffic is generated by web traffic. 
These measurements are reasonably consistent with (except for a 20% increase 
in web traffic) an earlier analysis of SIGCOMM 2001 conference traces covering 
4 APs and 195 stations. Bursty web traffic in our workload follows the self-similar 
ON-OFF traffic model given by Barford and Crovella (ACM SIGMETRICS 1998). 
We emulate the HTML transfer, browser processing, and HTML object retrieval phases 
with wget page retrieval tool. User's thinking time X between page accesses is 
Pareto distributed, with parameters k=1.5 and x_m=1, as suggested by Barford and 
Crovella. The remaining share of the workload comprises a mix of VoIP traffic 
(over UDP/IP) using the G.711 codec with H.323 signalling (3% of overall volume), 
and TCP packet transfers with exponentially distributed interarrival times 
(21% of the overall volume on average) as background traffic. These flows are 
emulated using the D-ITG traffic generator v.2.4.3.

Plain Text Files (UNIX CR/LF in GZip Compressed Tar)

In the provided trace collection, three files are included: paretoSamples.txt, 
objectSizes.txt, DITG_samples.txt. Pareto sample file is from one instantiation of 
our pareto random number generator, giving user thinking time values (over 500K values)
distributed between 15 and 100 (seconds). Object size file is a collection of 
the sizes (in bytes) of random web objects (from http://www.winlab.rutgers.edu website). 
These objects include plain html pages, html pages with images, and various file 
downloads (presentation files, source code tarballs, etc.). The way to generate 
WWW traffic using these traces should be as follows: Each node uses a disjoint 
part of the pareto sample file and requests a random object (of sizes listed 
in objectSizes.txt) by calling wget to a web server (e.g., Apache httpd, or MS IIS) 
that serves objects of these sizes. Then node goes to sleep (emulation of browsing 
and thinking) for an interval given by the next value in paretoSamples.txt file. 
Process continues with another random object from the list. The rest of the realistic 
traffic pattern (as explained above) is created by DITG and examples of how we created 
exponential and CBR type TCP traffic, and VoIP traffic are provided in DITG_samples.txt 
file.

[Trace] rutgers/ap_density/tcpdump/1AP_75STA (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-tcpdump-1AP_75STA-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/tcpdump/1AP_75STA (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/tcpdump/1AP_75STA},
  month = aug,  
  year = 2007
}
					

Traffic Dump for 1 AP 75 STA Experiment.

- Traffic Dump for 1 AP 75 STA Experiment

This file contains a frame dump file as seen from the wired port of 
the access point(s), for the application traffic travelling between 
wireless stations and application servers. The volume of this traffic 
is considered to be a metric for the performance of the WLAN under 
experimentation, since WLAN is deployed to haul traffic back and forth 
between wired and wireless networks. Trace starts at UNIX epoch 1168535241 
and finishes at 1168535821 (i.e. lasting approximately 10 minutes). 
These traces are from the experiments that use the realistic traffic 
workload as explained above.

TCPDump File (GZip Compressed Tar)

[Trace] rutgers/ap_density/tcpdump/2AP_75STA (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-tcpdump-2AP_75STA-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/tcpdump/2AP_75STA (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/tcpdump/2AP_75STA},
  month = aug,  
  year = 2007
}
					

Traffic Dump for 2 AP 75 STA Experiment.

- Traffic Dump for 2 AP 75 STA Experiment

This file contains a frame dump file as seen from the wired port of 
the access point(s), for the application traffic travelling between 
wireless stations and application servers. The volume of this traffic 
is considered to be a metric for the performance of the WLAN under 
experimentation, since WLAN is deployed to haul traffic back and forth 
between wired and wireless networks. Trace starts at UNIX epoch 1168540183 
and finishes at 1168540763 (i.e. lasting approximately 10 minutes). 
These traces are from the experiments that use the realistic traffic 
workload as explained above.

TCPDump File (GZip Compressed Tar)

[Trace] rutgers/ap_density/tcpdump/3AP_75STA (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-tcpdump-3AP_75STA-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/tcpdump/3AP_75STA (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/tcpdump/3AP_75STA},
  month = aug,  
  year = 2007
}
					

Traffic Dump for 3 AP 75 STA Experiment.

- Traffic Dump for 3 AP 75 STA Experiment

This file contains a frame dump file as seen from the wired port of 
the access point(s), for the application traffic travelling between 
wireless stations and application servers. The volume of this traffic 
is considered to be a metric for the performance of the WLAN under 
experimentation, since WLAN is deployed to haul traffic back and forth 
between wired and wireless networks. Trace starts at UNIX epoch 1168542112 
and finishes at 1168542752 (i.e. lasting approximately 10 minutes). 
These traces are from the experiments that use the realistic traffic 
workload as explained above.

TCPDump File (GZip Compressed Tar)

[Trace] rutgers/ap_density/tcpdump/4AP_75STA (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-tcpdump-4AP_75STA-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/tcpdump/4AP_75STA (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/tcpdump/4AP_75STA},
  month = aug,  
  year = 2007
}
					

Traffic Dump for 4 AP 75 STA Experiment.

- Traffic Dump for 4 AP 75 STA Experiment

This file contains a frame dump file as seen from the wired port of 
the access point(s), for the application traffic travelling between 
wireless stations and application servers. The volume of this traffic 
is considered to be a metric for the performance of the WLAN under 
experimentation, since WLAN is deployed to haul traffic back and forth 
between wired and wireless networks. Trace starts at UNIX epoch 1168543425 
and finishes at 1168544065 (i.e. lasting approximately 10 minutes). 
These traces are from the experiments that use the realistic traffic 
workload as explained above.

TCPDump File (GZip Compressed Tar)

[Trace] rutgers/ap_density/tcpdump/1AP_4STA (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-tcpdump-1AP_4STA-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/tcpdump/1AP_4STA (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/tcpdump/1AP_4STA},
  month = aug,  
  year = 2007
}
					

Wireless Traffic Dump for 1 AP 4 STA Experiment.

- Wireless Traffic Dump for 1 AP 4 STA Experiment

This file contains a frame dump file as seen by a wireless sniffer placed 
in the close vicinity of the WLAN being tested. There is no other process 
on the wireless sniffer and this trace is later used to derive PER statistics 
due to collisions together with driver packet error statistics from stations 
and AP(s). Due to the nature of the experiment, workload is constructed 
from clients to a central server by using 1300-byte frames sent using 
TCP at the maximum possible pace TCP allows. Trace lasts about five minutes.

TCPDump File (GZip Compressed Tar)

[Trace] rutgers/ap_density/tcpdump/1AP_8STA (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-tcpdump-1AP_8STA-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/tcpdump/1AP_8STA (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/tcpdump/1AP_8STA},
  month = aug,  
  year = 2007
}
					

Wireless Traffic Dump for 1 AP 8 STA Experiment.

- Wireless Traffic Dump for 1 AP 8 STA Experiment

This file contains a frame dump file as seen by a wireless sniffer placed 
in the close vicinity of the WLAN being tested. There is no other process 
on the wireless sniffer and this trace is later used to derive PER statistics 
due to collisions together with driver packet error statistics from stations 
and AP(s). Due to the nature of the experiment, workload is constructed 
from clients to a central server by using 1300-byte frames sent using 
TCP at the maximum possible pace TCP allows. Trace lasts about five minutes.

TCPDump File (GZip Compressed Tar)

[Trace] rutgers/ap_density/tcpdump/1AP_16STA (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-tcpdump-1AP_16STA-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/tcpdump/1AP_16STA (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/tcpdump/1AP_16STA},
  month = aug,  
  year = 2007
}
					

Wireless Traffic Dump for 1 AP 16 STA Experiment.

- Wireless Traffic Dump for 1 AP 16 STA Experiment

This file contains a frame dump file as seen by a wireless sniffer placed 
in the close vicinity of the WLAN being tested. There is no other process 
on the wireless sniffer and this trace is later used to derive PER statistics 
due to collisions together with driver packet error statistics from stations 
and AP(s). Due to the nature of the experiment, workload is constructed 
from clients to a central server by using 1300-byte frames sent using 
TCP at the maximum possible pace TCP allows. Trace lasts about five minutes.

TCPDump File (GZip Compressed Tar)

[Trace] rutgers/ap_density/tcpdump/2AP_4STA (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-tcpdump-2AP_4STA-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/tcpdump/2AP_4STA (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/tcpdump/2AP_4STA},
  month = aug,  
  year = 2007
}
					

Wireless Traffic Dump for 2 AP 4 STA Experiment.

- Wireless Traffic Dump for 2 AP 4 STA Experiment

This file contains a frame dump file as seen by a wireless sniffer placed 
in the close vicinity of the WLAN being tested. There is no other process 
on the wireless sniffer and this trace is later used to derive PER statistics 
due to collisions together with driver packet error statistics from stations 
and AP(s). Due to the nature of the experiment, workload is constructed 
from clients to a central server by using 1300-byte frames sent using 
TCP at the maximum possible pace TCP allows. Trace lasts about five minutes.

TCPDump File (GZip Compressed Tar)

[Trace] rutgers/ap_density/tcpdump/2AP_8STA (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-tcpdump-2AP_8STA-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/tcpdump/2AP_8STA (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/tcpdump/2AP_8STA},
  month = aug,  
  year = 2007
}
					

Wireless Traffic Dump for 2 AP 8 STA Experiment.

- Wireless Traffic Dump for 2 AP 8 STA Experiment

This file contains a frame dump file as seen by a wireless sniffer placed 
in the close vicinity of the WLAN being tested. There is no other process 
on the wireless sniffer and this trace is later used to derive PER statistics 
due to collisions together with driver packet error statistics from stations 
and AP(s). Due to the nature of the experiment, workload is constructed 
from clients to a central server by using 1300-byte frames sent using 
TCP at the maximum possible pace TCP allows. Trace lasts about five minutes.

TCPDump File (GZip Compressed Tar)

[Trace] rutgers/ap_density/tcpdump/2AP_16STA (v. 2007-08-09)

top

the initial version

@MISC{rutgers-ap_density-tcpdump-2AP_16STA-2007-08-09,
  author = {Mesut Ali Ergin and Kishore Ramachandran and Marco Gruteser},
  title = {{CRAWDAD} trace rutgers/ap_density/tcpdump/2AP_16STA (v. 2007-08-09)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/rutgers/ap_density/tcpdump/2AP_16STA},
  month = aug,  
  year = 2007
}
					

Wireless Traffic Dump for 2 AP 16 STA Experiment.

- Wireless Traffic Dump for 2 AP 16 STA Experiment

This file contains a frame dump file as seen by a wireless sniffer placed 
in the close vicinity of the WLAN being tested. There is no other process 
on the wireless sniffer and this trace is later used to derive PER statistics 
due to collisions together with driver packet error statistics from stations 
and AP(s). Due to the nature of the experiment, workload is constructed 
from clients to a central server by using 1300-byte frames sent using 
TCP at the maximum possible pace TCP allows. Trace lasts about five minutes.

TCPDump File (GZip Compressed Tar)

[Author] Mesut Ali Ergin

top

[Author] Kishore Ramachandran

top

[Author] Marco Gruteser

top

[Paper] ergin-density

top

In this paper, we present a systematic experimental study of the effect of 
inter-cell interference on IEEE 802.11 performance. With increasing penetration 
of WiFi into residential areas and usage in ad hoc conference settings, chaotic 
unplanned deployments are becoming the norm rather than an exception. These 
networks often operate many nearby access points and stations on the same 
channel, either due to lack of coordination or insufficient available channels. 
Thus, inter-cell interference is common but not well-understood. According to 
conventional wisdom, the efficiency of an 802.11 network is determined by the 
number of active clients. Surprisingly, we find that with a typical 
TCP-dominant workload, cumulative system throughput is characterized by the 
number of interfering access points rather than the number of clients. We find 
that due to TCP flow control, the number of backlogged stations in such a 
network equals twice the number of access points. Thus, a single access point 
network proved very robust even with over one hundred clients. Multiple 
interfering access points, however, lead to an increase in collisions that 
reduces throughput and affects volume of traffic in the network.