2012 9th Annual Conference on Wireless On-Demand Network Systems and Services (WONS)
WiFi network residual bandwidth estimation:
a prototype implementation
Andrea Ghittino, Nazario Di Maio, Domenico Di Tommaso
CSP Inno vazione nelle ICT
Andrea.gh email@example.com, firstname.lastname@example.org, do email@example.com
Abstract— Evaluating the available bandwidth in a wireless
LAN is a challenging task because the throughput depends on
several factors such as RSSI (Received Signal Strength
Indication), interf erer s and packet size. Moreover, in a real
An important step for performance measurement is the
scenario it is not sufficient to evaluate periodic ally the
throughput of the network, but a continuous monitoring is
required in order to detect as soon as possible network
problems and saturation phenomena. When wireless LAN is
used to transmit high priority traffic (e.g. voice and video
flows), an admission control procedure based on residual
bandwidth evaluation is necessary to verify that required
resources are available.
We propose a mechanism based on active traffic probing
with low priority packets: this approach enables us to simulate
a real data flow and, on the other hand, does not interfere with
existing flows. Our method relies on IEEE 802.11e (WMM -
Wireless MultiMedia) support and doe s not require any
customization of network devices.
Through this approach the residual bandwidth can be
continuously monitored. This measurement accounts for both
the characteristics of active flows (different pac ket sizes and
rates) and external int erferenc es, offering accurate bandwidth
As wireless networks are getting more and more
widespread, there is a big interest around the topic of
available bandwidth estimation methods. This argument
poses a serious challenge for a shared medium with rate
adaptation such as the IEEE 802.11 networks. In particular,
in a scenario with QoS (Quality of Service) traffic
classification, the information about residual bandwidth
could be very useful, especially in the admission control
An analysis of the state of the art of residual capacity
estimation algorithms highlights the presence of two main
approaches: active and passive. In the first one, probe
packets are used to infer the current state of the network. The
latter is a non-intrusive method, as no additio nal packets are
inserted into the system. This method is essentially based on
listening channel activity to estimate the channel idleness
choice of the right estimation technique and whether that
method is appropriate for our particular network
enviro nment. There are situations in which it is possible to
carry out performance tests on an unused network: in this
case a certain amount of test traffic is transmitted from a
source to a destination until the network exhibits losses.
More often, it is necessary to measure the performance of
a working WLAN; in such a situation, it is possible that
many users are currently utilizing the network and the test
must be carried out transparently, disturbing the
communication channel as less as possible. Hence, we can
mention at least two different approaches:
•active techniques. These techniques rely on the
emission of dedicated end-to-end probe packets to
estimate the available band width along a path, as
described in .
•passive techniques. These techniques use only local
information on the utilization of the bandwidth. A
typical example of such approaches is a node
monitoring the channel usage by sensing the radio
medium. These mechanisms are usually
transparent, but they may rely on exchanging
information with neighbors via one-hop broadcast
To measure network characteristics, such as the available
band width, the use of passive measurement methods is a
possible strategy. Passive measurement methods and tools
act as observers inside a network and usually they will not
interfere with other traffic.
Instead of using passive observers as described abo ve, we
can deploy active measurements methods. For example,
these methods include the injection of so-called probe traffic
into the network at a traffic source, and calculate the end-to-
end available bandwidth by measuring the one-way delays of
these probing packets.
Hence, active measurement methods without further
expedients, affect the network traffic itself.
The rest of the paper is organized as follows. In section II
we review previous work on this subject, while in section III
978-1-4577-1722-2/12/$26.00 ©2012 IEEE
we describe our solution and in section IV we present our
preliminary results. Section V draws some conclusions and
outlines future work.
II. RELATED WORK
Many tools are available to evaluate the throughput on a
specific wireless interface:  co mpares active and passive
tools to evaluate the performance of a wireless netwo rk with
varying interference levels and data rates. However, active
as a test client (i.e. the transmitting node), it begins
the measurement procedure by generating the
probing traffic to wards the “receiving” node;
as a test server (i.e. the receiving node), it keeps
track of the incoming traffic, evaluates the
effective residual band width on the basis o f the
average throughput reached during the session and
eventually delivers the results to the controller.
tools are used to evaluate the overall throughp ut without
considering active user flows.
Moreover, several researchers have studied the problem
of bandwidth estimation in a wireless LAN. In  a hybrid
method based on active and passive approach is suggested.
Authors affirm that the active method, based on traffic flow
generation to emulate user data exchange, is accurate, but the
main drawback is that these flows compete with real data
from the users.
In , a method based on sending two back-to-back
probe packets to a neighb or is considered. Dispersion of
packets is measured. The mechanism relies on WMM: the
first probe packet uses a high-priority queue, and the second
probe uses a low-priority queue. To evaluate the throughput,
authors exploited 802.11e features to obtain a tighter control
over MAC activity, achieving a resolution of the order of
III. THE PROPOSED SOLUTION
A. Our architecture
The implemented solution has been chosen taking into
account some possible future applications, with a special
attention to the distribution of multimedia flows in a
domestic environment. Our application is suited for a SOHO
environment in which QoS must be provided. Our
application can rely on commercial WiFi routers, if they are
The aim of our tool is to set up a test session every time it
is required to estimate the residual available bandwidth
between two specific nodes. From a general point of view,
the developed system consists of a controller entity and
several daemons (background services) running at network
nodes. Eventually, the two entities can both coexist in a
single network node.
The first one is in charge of the overall management
procedure, including setting the system parameters in order
to configure the single test sessions and retrieve the test
results, once they are available. The latter performs a
peculiar set of tasks, such as handling the signaling protocol
inputs and executing the measurement routine itself.
Once all daemons have received the instructions and
parameters for the test session, the controller starts the
estimation procedure. At this stage, the daemon behaves in
different ways depending on its designated role during the
As mentioned above, signaling is need ed for node
coordination during the test set up and to avoid that
simultaneous evaluation sessions take place in the same
wireless channel cluster. In fact, this would compromise the
results, as explained in the following paragraph.
The signaling protocol is based on a set of primitives,
each of which triggers a specific component of the
measurement sequence in the daemons:
The INIT procedure is the first protocol step: the two
daemons selected fo r residual bandwidth estimation listen for
the configuration parameters that will be sent by the
contro ller. In this way, the controller is allowed to schedule a
seq uence of tests with different target nodes at different
times, even in a continuous cycle. The controlling parameters
•IP addresses of the test nodes;
•ports for signaling and probe traffic;
•node role (server or client for probing)
•protocol for probing data flow (TCP/UDP);
•number of packets;
•probe traffic throughput required;
•test procedure duration;
•probe traffic DSCP (Differentiated Services Code
The choice of these parameters influences results. Hence,
it is important to choose them accurately.
For example, the packet size is a key parameter for the
UDP protocol; on the other hand, for the TCP transport
protocol efficiency, the test duratio n is a critical parameter.
Longer sessions generate more reliable and trustworthy
measurements, but in a real-time context it could be crucial
to obtain a quick evaluation of the available bandwidth. The
DSCP parameter identifies the class used by the probe to tag
the packets and it is used to identify the WMM queue.
Once the set-up of the test is completed, the controller
can send the START signal to the two daemons in charge of
the test. This step has the only purpo se of synchro nizing the
involved entities in order to assure that the server node is
ready to receive the test traffic flow from the client node.
Using the GET RESULTS primitive, which is the last
pro tocol step, the controller asks the server node to supply a
results report. Although the test client node keeps track of
the test and knows the transmission throughput reached
(possibly the same asked by the controller), the server is the
only one that measured the exact amount of received data.
The report co ntains information about:
The last value, together with the test duration, gives us
the estimated value of residual channel capacity averaged
over the time interval the test has been performed.
B. Implementa tion Features
One of the key concepts behind the development of our
evaluation tool is that probe traffic generation and injection
in the network must serve as a measuring tool. Then, it must
not affect in any way the pre-existing traffic.
This is the reason why WMM has been employed. The
generated prob e traffic is marked as “background” in order
to have the lowest priority both in the internal virtual
transmission queue of the node and in the external queue in
the channel access phase.
This is implemented using QoS traffic tagging as
specified by the WMM access categories .
enviro nment such as a wireless network for multimedia
entertainment applications (IEEE802.11g).
The following table shows the maximum throughput we
reached out of theoretical 54Mbps in the communication
between an access point and a station under different type of
traffic. Obviously, UDP probe traffic is characterized by
lower rate d ue to disadvantageous WMM parameters.
Table 2: maximum throughput for different data type.
First co nsideration can be done comparing the graphs in
which is represented UDP and TCP data flow that saturate
the wireless channel. The UDP probe traffic, has worse
WMM parameters with respect to the data traffic. Figure 1
and Figure 2 show that the residual bandwidth estimation do
not affect user data flow (UDP/TCP) going to ~0 Mbps when
saturating data traffic occurs.
The probing traffic transparency is obtained thanks to the
high difference in WMM parameters which corresponds to
not efficient exp loitation of channel access timings by
This behavior is deterministic and can be inferred
looking at Table 2. Hence, we have to sum the predicted gap
to the obtained esteem in order to find out the real available
bandwidth taking into account that such gap depends on the
data traffic protocol and WMM parameters assigned to it.
Table 1: parameters caracterising the WMM QoS access categories of a
Wi-Fi access point and station (station traffic values in the second rows of
each table cell).
For generating traffic and measuring, our deamon relies
on iperf3. It is a widely used network testing tool that can
generate TCP and UDP data streams over a network and
measure its throughput.
Before you begin to format your paper, first write and
save the content as a separate text file. Keep your text and
graphic files separate until after the text has been formatted
and styled. Do not use hard tabs, and limit use of hard returns
to only one return at the end of a paragraph. Do not add any
kind of pagination anywhere in the paper. Do not number
text heads-the template will do that for you.
Finally, complete content and organizational editing
before formatting. Please take note of the following items
when proofreading sp elling and grammar:
Figure 1: TCP data traffic with Probing Traffic.
available bandwidth estimation gives different results: for
small packet size, resid ual channel capacity decreases, due
to the large amount of channel access attempts by data
traffic. Hence, the estimation takes into account the previous
highlighted variables and the scenario conditions.
Figure 2: UDP data traffic with Probing Traffic.
Figure 3: UDP data traffic (datagram of 64Byte) with Probing Traffic.
Figure 4: UDP data traffic (datagram of 256Byte) with Probing Traffic.
Figure 3,4,5 show sessions of data flow with differen t
datagram lengths (64, 256, 512 Bytes). It is worth pointing
out that, even if data traffic is always the same (2Mb ps), the
Figure 5: UDP data traffic (datagram of 512Byte) with Probing Traffic.
V. CONCLUSION AND FUTURE WORKS
In this paper we presented the design and the first
implementation of a bandwidth estimator based on active
traffic generation. This approach needs router Wireless
Multimedia Extension compliant avoiding negative effects
degradation on user data flows. We assessed the correct
behavior of the system in a simple scenario consisting of one
access point and several stations. We are planning to
analyze the impact of WME parameters on the performance
in order to accurately evaluate the error in the residual
band width estimation. Moreover, we also plan to verify our
tool on Wireless Internet Service Provider (WISP) networks
that are typically based on 5GHz links. In this scenario,
accurate bandwidth estimation could improve network
planning and design of new links. We also plan to extend our
tool and use it in wireless mesh routing protocols as a new
metric to calculate the best path.
 C.Sarr, C.Chaudet, G.Chelius, I.G.Lassous, “Available Bandwidth
Estimation for IEEE802.11-based Ad Hoc Networks”, IEEE
Transactions On Mobile Computing, October 2008.
 D.Gupta, D.Wu, P.Mohapatra, C.Chuah, “”Experimental Comparison
of Bandwidth Estimation Tools for Wireless Mesh Networks”, In
Proc. INFOCOM 2009, IEEE, Rio de Janeiro, Brazil, 19-25 April,
 B.Landfeldt, P.Sookavatana, A.Seneviratne, “The case for a hybrid
passive/active network monitoring scheme in the wireless Internet”,
In Proc. ICON 2000, 5-8 September, 2000.
 M.A.Ergin, M. Gruteser, “Using Packet Probes for Available
Bandwidth Estimation: A Wireless testbed Experience”, In Proc.
Wintech 2006, Los Angeles, California, USA, 29 Semptember, 2006.
 Wi-Fi Alliance, “Wi-Fi CERTIFIED for WMM – Support for
Multimedia Applications with Quality of Service in Wi-Fi
Networks”, White Paper, September, 2004.