|
Overview
|
- A brief description
of the project
|
|
Data
|
- Parameters for the derived
flow models
|
|
Documents
|
- Papers and
presentation slides
|
|
People
|
- The people behind the
project
|
|
Related
Reseach
|
- Relevant projects to
check out
|
Overview
This project uses real traces to model
and understand the behavior of data flows in large-scale wireless LANs. Data
flows, namely, TCP connections, are essential to most of the applications running
on wireless LAN networks. Flows are also important to many protocol designs
and performance evaluation issues, e.g., TCP congestion control, routing,
flow-level scheduling, etc. Therefore, an accurate flow-level model can be extremely
useful for both the research community and network operator.
In this project, we statistically
characterize both static flows and roaming flows in a large campus wireless
network using real traces collected by the research community.
·
Static flows. A static flow is defined as a flow which uses only one Access
Point (AP) within its duration. A two-tier approach is used for modeling static
flow traffic at individual Access Point, which results a Weibull regression
model. An observation is that the static flow arrivals in spatial proximity
show strong similarity.
·
Roaming flows. A roaming flow is defined as a flow during whose duration more than
one AP is used. The population of roaming flows is very small (0.13%). They
can also be characterized by residing time and handoff frequency.
In addition to modeling, the project
discusses the physical explanations, particularly user behaviors, behind
the statistics.
The accuracy of flow models is very important.
To illustrate this, we use two examples: TCP evaluation and flow-level
scheduling algorithms.
Data
In the following, we provide the
parameters for flow models. These parameters are based on the Spring 2002
data collected from Dartmouth College.
·
Flow arrival traffic at AP. The flow arrival traffic at 15 APs are studied. For
each AP, the flow arrival traffic are modeled by flow inter-arrival time
and flow data size.
o Flow
inter-arrival time. Each day is
divided into 24 hours. The flow inter-arrival time in each hourly interval
is modeled by Weibull model (Details about Weibull model is referred to our
paper). All the parameters are provided in InterarrivalTime.txt.
In this file, each row represents an AP. The
format for each row is as following:
time_of_day Weibull_para_1,
Weibull_para_2 …
o Flow
data size. Flow data size is modeled
by the Lognormal distribution. Datasize.txt provides
the parameters. Again, each row represents an AP. The format for each row
is as following:
Lognormal_para_1, Lognormal_para_2 …
·
Roaming flows. Roaming flows are modeled by two metrics: handoff frequency and
residing time.
o Handoff
frequency: number of handoff
within the flow duration. This metric is modeled by Geometric distribution.
The parameter in the Geometric distribution is 0.80.
o Residing
time: within the duration of a
roaming flow, the time that the user is associated with an AP. This metric
is modeled by the Weibull distribution. The two parameters are 0.1491, 0.3770
respectively.
Documents
- " Characterizing
Flows in Large Wireless Data Networks ", Xiaoqiao (George) Meng,
H.Y. Starsky Wong, Yuan Yuan, Songwu Lu, In Proc. of ACM MOBICOM 2004
[pdf]
- "Characterizing
Flows in Large Wireless Data Networks", presented at ACM MOBICOM,
Philadelphi, Sep. 2004 [ppt]
Project Members
Faculty
Students
Related Research
|
Home
Wireless QoS
People
Publications
Software
Talks
Support
Links