[Dataset] stanford/gates (v. 2003-10-16)

top

The initial version

@MISC{stanford-gates-2003-10-16,
  author = {Diane Tang and Mary Baker},
  title = {{CRAWDAD} data set stanford/gates (v. 2003-10-16)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/stanford/gates},
  month = oct,  
  year = 2003
}
					

This dataset contains traces of the Stanford CS department's wireless network.

We collected a 12-week trace of a local-area wireless network installed 
throughout the Gates Computer Science Building of Stanford University.

The building is L-shaped (the longer edge is called the a-wing, and the
shorter the b-wing). It has four main floors with offices and labs, a
basement with classrooms and labs, and a fifth floor with a lounge
and a few offices. Each of the main floors has two access points,
one for each wing. Additionally, the first floor has an access point
for a large conference room; the library, which spans both the
second and third floors, also has an access point. The basement
has two access points, one near the classrooms and one for the
Interactive Room, a special research project in the department.
The smaller fifth floor only has one access point.
The wireless user community consists of 74 users who can be
roughly divided into four groups:

- 35 first year PhD students, who were each given a laptop
with a WaveLAN card upon arrival (which corresponds to
the beginning of the trace). Their offices are primarily in the
2b wing.
- 22 graphics students and staff, the majority of whom
received laptops with WaveLAN cards a week into the
tracing period. Their offices are primarily in the 3b wing.
- Three robots, used by the robotics lab for research. The
robots do not have to authenticate themselves to reach the
outside network. While the robots are somewhat mobile, they
stay in the 1a wing. Although these WaveLAN cards are
intended to be used by the robots, students in the robotics lab
also use the network cards for session connections and websurfing.
- 14 other users (students, staff, and faculty) scattered
throughout the building.

In addition to these 74 users, there were also four users who
authenticated themselves but only connected to wired ports on the
public subnet rather than the wireless network. We do not
consider these users in the rest of this analysis of the wireless
network.

In the Gates Computer Science Building at Stanford University, 
administrators have made a "public" subnet available for any user 
affiliated with the university. Users desiring network access     
via this subnet must authenticate themselves to use their dynamically 
assigned IP address to access the rest of the departmental and 
university networks and the Internet.
This subnet is accessible both from a wireless network and from 
Ethernet ports in public places in the building, such as conference rooms,
lounges, the library, and labs. The wireless network is a
WaveLAN network with WavePoint II access points acting as
bridges between the wireless and wired networks. The access
points each have two slots for wireless network interfaces; both
slots are filled, one with older 2 Mbps cards to support the few
users who have not updated their hardware yet, and the other with
WaveLAN IEEE802.11-compatible 10 Mbps cards.

Because all of the wireless users are on a single subnet
(which promotes roaming without the need for Mobile IP or other
such support), we gathered traces on the router that connects 
the public subnet to the rest of the departmental
wired network. The router is a 90 MHz Pentium running RedHat
Linux with two 10 Mbps network interfaces. One interface
connects to the public subnet, and the other connects to the
departmental network.

To gather all of the information we wanted, we collected
three separate types of traces during a 12-week period
encompassing the 1999 Fall quarter (from Monday, September 20
through Sunday, December 12). 

The first trace we gathered is a tcpdump trace of the link-level 
and network-level headers of all packets that went through the router. 
We use this information in conjunction with the other two traces.

The second trace is an SNMP trace. Approximately every
two minutes, the router queries, via Ethernet, all twelve access
points for the MAC addresses of the hosts currently using that
access point as a bridge to the wired network. Once we know
which access point a MAC address uses for network access, we
know the approximate location (floor and wing) of the device with
that MAC address. We pair these MAC addresses with the link
level addresses saved in the packet headers to determine the
approximate locations of the hosts in the tcpdump trace.
The overhead from the SNMP tracing is low: 530 packets or
50 KBytes is the average overhead from querying all twelve
access points every two minutes. The overhead for querying an
individual access point is 3.2 KBytes if no MAC addresses are
using that access point; otherwise, the base overhead is 14.5
KBytes for one user at an access point, plus 1 KByte for every
additional user.

The last trace is the authentication log, which keeps track of
which users request authentication to use the network. Each
request has both the user's login name as well as the MAC address
from which the user makes the request. We pair these MAC
addresses with the link-level addresses saved in the tcpdump trace
to determine which user sends out each packet.

We obtained permission to collect these traces from the
Department Chair and informed all network users that this tracing
was taking place. We additionally informed users we would
record packet header information only (not the contents) and that
we would anonymize the data. Knowledge of the tracing may have
perturbed user behavior, but we have no way of quantifying the
effect.

[Traceset] stanford/gates/combined (v. 2003-10-16)

top

The initial version

@MISC{stanford-gates-combined-2003-10-16,
  author = {Diane Tang and Mary Baker},
  title = {{CRAWDAD} trace set stanford/gates/combined (v. 2003-10-16)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/stanford/gates/combined},
  month = oct,  
  year = 2003
}
					

This traceset contains traces of the Stanford CS department's wireless network.

We use the common timestamp and MAC address
information to combine three traces 
(tcpdump, SNMP, and authentication logs) into a single trace. 
The original three traces are not publicly available.

We have anonymized the user and remote host names 
for privacy reasons.

[Trace] stanford/gates/combined/anon (v. 2003-10-16)

top

The initial version

@MISC{stanford-gates-combined-anon-2003-10-16,
  author = {Diane Tang and Mary Baker},
  title = {{CRAWDAD} trace stanford/gates/combined/anon (v. 2003-10-16)}, 
  howpublished = {Downloaded from http://crawdad.cs.dartmouth.edu/stanford/gates/combined/anon},
  month = oct,  
  year = 2003
}
					

This trace contains traces of the Stanford CS department's wireless network.

[time] [pkt size] [username] [access point loc] [app] [dir] [remote host] 

dir is the direction -- incoming or outgoing or both 
(i.e., internal, or neither i.e., 
dhcp hadn't really gotten its act together yet). 
app will be a dotted port number (src/dst) 
if it's not recognized. 
time is at second granularity. 
pkt size is in bytes.

We use the common timestamp and MAC address
information to combine these three traces 	
(tcpdump, SNMP, and authentication logs) into a single 
trace with a total of 78,739,933 packets attributable 
to the 74 wireless users.
An additional 37,893,656 packets are attributable to 
the SNMP queries and 1,551,167 packets are attributable 
to the four wired users. The number of packets attributable 
to the SNMP queries might seem high, but each access point 
is queried every two minutes even if no laptops are 
actively generating traffic.

Note that because we do not record any signal strength
information, and since our access points generally cover 
a whole wing of a floor, we cannot necessarily detect 
movement within a wing but only movement between access points.

[Author] Diane Tang

top

[Author] Mary Baker

top

[Paper] bahl-breathing

top

[Paper] meng-flows

top

Several studies have recently been performed on wireless university campus 
networks, corporate and public networks. Yet little is known about the 
flow-level characterization in such networks. In this paper, we statistically 
characterize wireless network using a recently-collected trace. For static 
flows, we take a two-tier approach to characterizing the flow arrivals, which 
results a Weibull regression model. We further discover that the static flow 
arrivals in spatial proximity show strong similarity. As for roaming flows, 
they can also be well characterized statistically. We explain the results by 
user behaviors and application demands, and further crossvalidate the modeling 
results by three other traces. Finally, we use two examples to illustrate how 
to apply our models for performance evaluation in the wireless context.

[Paper] tang-wavelan

top

To understand better how users take advantage of wireless networks, we examine 
a twelve-week trace of a building-wide local-area wireless network. We analyze 
the network for overall user behavior (when and how intensively people use the 
network and how much they move around), overall network traffic and load 
characteristics (observed throughput and symmetry of incoming and outgoing 
traffic), and traffic characteristics from a user point of view (observed mix 
of applications and number of hosts connected to by users). Amongst other 
results, we find that users are divided into distinct location-based 
sub-communities, each with its own movement, activity, and usage 
characteristics. Most users exploit the network for web-surfing, 
session-oriented activities and chat-oriented activities. The high number of 
chat-oriented activities shows that many users take advantage of the mobile 
network for for synchronous communication with others. In addition to these 
user-specific results, we find that peak throughput is usually caused by a 
single user and application. Also, while incoming traffic dominates outgoing 
traffic overall, the opposite tends to be true during periods of peak 
throughput, implying that significant asymmetry in network capacity could be 
undesirable for our users. While these results are only valid for this 
local-area wireless network and user community, we believe that similar 
environments may exhibit similar behavior and trends. We hope that our 
observations will contribute to a growing understanding of mobile user 
behavior.