ENHANCEMENT OF TCP FAIRNESS IN IEEE 802.11 NETWORKS

Natarajan Meghanathan, et al. (Eds): SIPM, FCST, ITCA, WSE, ACSIT, CS & IT 06, pp. 357–365, 2012.
© CS & IT-CSCP 2012 DOI : 10.5121/csit.2012.2335
ENHANCEMENT OF TCP FAIRNESS IN IEEE
802.11 NETWORKS
Bhuvana LN1
, Divya PD2
and Divya A3
Department of Information Technology, Easwari Engineering College,
Affiliated to Anna University, Chennai
1
bhuvana.n21@gmail.com
2
div.pd91@gmail.com
3
divi_cool8@yahoo.com
ABSTRACT
The usage of fixed buffers in 802.11 networks has a number of disadvantages associated with
it. This includes high delay, reduced throughput and inefficient channel utilisation. To
overcome this, a dynamic buffer sizing algorithm, the A* algorithm has been implemented at
the access point. In this algorithm buffer size is dynamically adjusted depending upon the
current channel conditions and hence delay is reduced and the throughput is maintained. But
in 802.11 networks with DCF collision avoidance mechanism, it creates significant amount of
unfairness between the upstream and downstream TCP flows, with clusters of upstream ACKs
blocking downstream data at the access point. Thus a variation of the Explicit Window
Adaptation (EWA) scheme has been used to regulate the queuing time of the upload clients by
calculating the feedback value at the access point. This creates fairness and increases the
number of transmission opportunities for the downstream traffic.
KEYWORDS
Wireless LANs(WLANs), Medium access control (MAC), Transmission control protocol
(TCP), Dynamic Buffer sizing, Explicit Window Adaptation(EWA).
1. INTRODUCTION
In computer networks, buffers are generally used to store short term packet bursts thereby
reducing packet drops and maintaining high link efficiency. Packets are also queued in buffers
when the network device lacks the ability to process all the packets immediately.
There are two queuing disciplines namely First In First Out/First Come First Serve (FIFO/FCFS)
that can be used by the buffer to queue the packets. In FIFO, the first packet that arrives at the
buffer of a network device say for example access point, will be the first packet to be transmitted.
When the number of packets exceeds the buffer size the packet at the tail is dropped. This
dropping policy is called the tail drop policy. The main disadvantage of FIFO is that it does not
discriminate between different traffic sources. It thus led to the introduction of Fair Queuing
technique. In this a separate queue is maintained for each flow currently handled by the network
device. The network device services these queues in a round robin fashion. Overflowing packets
belonging to each queue are discarded. An enhancement of Fair Queuing technique is the
Weighted Fair Queue, which assigns a weight to each queue serviced by the network device. The
weight defines the number of bits to transmit each time the device services the queue which in
turn affects the percentage of links bandwidth the queue will get. The general rule for sizing

358 Computer Science & Information Technology ( CS & IT )
buffers is to set the buffer size to be the product of bandwidth and delay of the flows utilizing the
link. This rule is known as the Bandwidth-Delay Product (BDP) rule.
When fixed buffers are used in 802.11 networks it leads to undesirable channel utilization and
high delay. The use of dynamic buffer increases the throughput and reduces the delay. In contrast
to wired networks the transmission in wireless networks are broadcast in nature. Thus the
buffering requirement at each station depends upon the number of active station in wireless LAN.
In dynamic buffer sizing techniques the size of buffers is changed to adapt to the changing
conditions of the network.
The A* algorithm which is a combination of eBDP and Adaptive Limit Tuning Feedback
algorithms (covered in the later sections) is used to change the size of buffers dynamically. While
the A* algorithm decreases delay and maintains the throughput as achieved by the use of fixed
buffers there is a disadvantage associated with this algorithm. The A* algorithm creates
significant amount of unfairness between the upstream and downstream TCP flows. The upstream
flow gets a greater share of the links available bandwidth when compared to the downstream flow
which gets a very small percentage of the total available bandwidth.
In our project a variation of the Explicit Window Adaptation (EWA) scheme which modifies the
Advertised Receiver Window field value of ACK packets at the Access Point is used to create
fairness among the different flows.
2. IEEE 802.11 MAC ALGORITHM
The Medium Access Control (MAC) Algorithm used by the 802.11 networks is the Distributed
Coordination Function (DCF) which is a CSMA/CA based algorithm. According to DCF, when
the wireless medium is idle for DIFS (Distributed Inter Frame Spacing) time period, each station
initializes a back off counter to a random number selected uniformly between 0 and CW-1 where
CW is the Contention Window size. The time period is slotted and the back off counter is
decremented each slot the medium is idle. If the medium is detected as busy the countdown halts
and the countdown resumes only when the medium is idle again for DIFS time period. When the
counter reaches zero a station transmits packet. If collision occurs the size of Contention Window
is doubled and the process gets repeated. When the transmission is successful CW is reduced to
CWmin and a new countdown is started.
3. EMULATING BDP ALGORITHM
In wireless network a fixed value of BDP does not exist because the mean service time is time
varying in nature. The service time of each station is measured by observing the time between the
arrival of the packet at the network interface queue ts and the successful transmission of packet te.
On averaging the packet service time of all stations the mean packet service time Tserv is obtained.
The eBDP algorithm works as follows. The maximum queuing delay Tmax is defined. 1/Tserv
indicates the mean service rate. The buffer size is given as QeBDP = min (Tmax/Tserv, QeBDP
max) where
QeBDP
max is the maximum buffer size. For every packet that arrives into the queue the value of
QeBDP is calculated. If the queue occupancy is less than QeBDP the packet is put into queue else it is
dropped. For every outgoing packet the end service time te is recorded as the time at which the
MAC ACK arrives for the packet and the service time for each packet is calculated as Tserv = (1-
W)Tserv + W(te-ts) where W is the averaging parameter.
4. ADAPTIVE LIMIT TUNING FEEDBACK ALGORITHM
To maintain high link efficiency a station must have a packet to transmit whenever it wins a
transmission opportunity. Thus the time for which the station buffer is empty should be
minimised which can be achieved by maintaining the buffer size sufficiently large. But the use of
large buffers can lead to high queuing delays. Thus to ensure low delay the buffer size should be

Computer Science & Information Technology ( CS & IT ) 359
as small as possible. Thus the smallest buffer size that ensure high link utilisation should be
selected for operation. The following approach is used to ensure effective link utilisation. If the
buffer rarely empties the buffer size is decreased. On the contrary if the buffer remains empty for
a long period the buffer size is increased.
Input ts,te
Input QeBDP,qALT
Input Ru/Rd
Input feedback
value
Figure 1: The above figure shows the overall functioning of the wireless network system.
In this algorithm a queue occupancy threshold qthr is defined. The amount of time the queue
spends below this threshold and amount of time the queue spends above this threshold are
observed as ti(k) and tb(k) respectively where ti(k) indicates the idle period and tb(k) the busy
period. The time period t= ti(k)+tb(k). The minimum link utilisation is given by tb/(tb+ti) and q(k)
denotes the buffer size at the kth
observation interval. The buffer size is updated as
q(k+1)=q(k)+a1ti(k)-b1tb(k) where a1 and b1 are the design parameters. If a1ti(k)=b1tb(k) the buffer
size remains unchanged, if the idle time is larger i.e. a1ti(k)>b1tb(k) the buffer size is increased and
if the busy time is larger i.e. a1ti(k)<b1ti(k) the buffer size is decreased. The maximum buffer size
is denoted as qmax and the minimum buffer size is given as qmin. For every time period t, the idle
period is calculated and qALT is measured as qALT+a1ti-b1(t-ti) and this value is updated as qALT=
min(max(qALT,qmin),qmax) .
5. THE A* ALGORITHM
The A* algorithm is a hybrid algorithm that uses the mean service time to calculate QeBDP in the
eBDP algorithm and uses the idle/busy time to calculate the value of qALT in ALT algorithm and
sets the buffer size at access point as min(QeBDP,qALT). When the channel condition changes the
service time measured from eBDP algorithm is used to adjust the buffer size accordingly. It then
uses the ALT algorithm to set the buffer size more accurately.
DATA TRANSFER FROM
SERVER TO DOWNLOAD
CLIENTS THROUGH
ACCESSPOINT
FIFO QUEUEING OF
PACKETS IN ACCESS
POINT
MEASURE
SERVICE TIME
AND IDLE
BUSY TIME
SET OPTIMAL
BUFFER
(QUEUE) SIZE
USING A*
DCF INCLUDED IN
ABOVE SCENARIO
ANALYSE PERFORMANCE
IN PRESENCE OF UPLOADS
INSTABILITY WITH
MORE THAN 2 UPLOADS
NOTED
FAIRNESS ENHANCED BY
USING EWA(EXPLICIT
WINDOW ADAPTATION)
VARIATION

Figure2: Illustrates simulation graph obtained for Buffer Occupancy (pkts)
downloads in case of fixed buffers and A* algorithm.
Figure3: Illustrates simulation graph obtained for Delay(RTT) (ms)
in case of fix
Figure 4: Represents RTT (ms)
6. UNFAIRNESS BETWEEN
Consider a WLAN networks made up of n client stations.
flow each. The access point of the WLAN is responsible for transmittin
to the client. But these ACK may be dropped because of
the data packets get n/(n+1) share of
Computer Science & Information Technology ( CS & IT )
Fixed Buffer
A* Algorithm
simulation graph obtained for Buffer Occupancy (pkts) vs Buffer Size (pkts) for 5
downloads in case of fixed buffers and A* algorithm.
Fixed Buffer
A* Algorithm
Figure3: Illustrates simulation graph obtained for Delay(RTT) (ms) vs Buffer Size (pkts) for 5 downloads
in case of fixed buffers and A* algorithm.
Figure 4: Represents RTT (ms) vs Number of Uploads with A* algorithm.
ETWEEN UPSTREAM AND DOWNSTREAM TCP F
Consider a WLAN networks made up of n client stations. The client stations carry one upload
oint of the WLAN is responsible for transmitting the ACK packets back
But these ACK may be dropped because of the equal access nature of WLAN, since
share of transmission opportunities whereas the ACK packets that
vs Buffer Size (pkts) for 5
vs Buffer Size (pkts) for 5 downloads
FLOWS
The client stations carry one upload
g the ACK packets back
the equal access nature of WLAN, since
transmission opportunities whereas the ACK packets that

represent the downstream traffic gets only 1/(n+1) share. This creates significant amount of
unfairness between the upstream and downstream TCP flows.
7. UNDERSTANDING TCP FAIRNESS IN 802.11 NETWORK
In WLANs the mobile host (sender/receiver) access the network through the access point (AP).
IEEE 802.11 ensures equal access to the media for all host.
If there is one sender and remaining are receivers, the access point and sender gets equal access to
the medium and i.e. the sender gets half of the total available bandwidth and the remaining half of
the bandwidth is used by the access point. The bandwidth available for the access point is shared
equally by all receivers.
The equal access nature of medium access protocol results in unfairness. The throughput ratio is
given as the ratio between average TCP uplink throughput (Ru) and the average TCP downlink
throughput (Rd). Even when there is a single mobile sender(upstream flow) and a single mobile
receiver(downstream flow) the throughput ratio is 1.44 which indicates that the sender receives
1.44 times the receiver’s bandwidth.
½ of total ½ of total
bandwidth bandwidth
¼ of total bandwidth
Figure 5: Represents a network made up of one sender and two receivers.
The sender and the access point gets equal share of the channels available bandwidth.
Four regions of TCP unfairness based upon the buffer size at the Access Point are identified. In
the first region the buffer size at AP is greater than 84 packets and the throughput ratio is one.
This indicates both the upstream and downstream flow gets equal share of the available
bandwidth. This is because the buffer is large enough to accommodate the maximum receiver
window of both flows. In the second region the buffer size is between 42 and 84 and the
throughput ratio lies between 10 and 1. The third region corresponds to that region where the
access point buffer size lies between 6 and 42 and the throughput ratio varies between 9 and 12.
In the fourth region the buffer size is less than 6.
Case 1 : One upstream and several downstream flows
Figure 6: Shows One upstream and n downstream flows
In this case the throughput ratio is linear i.e. all the downstream flow share the same resources
while the throughput remains stable.
Case 2 : Equal number of upstream and downstream flows
SENDER AP RECEIVER1
RECEIVER2
SENDER AP
RECEIVER 1
RECEIVER 2
RECEIVER n

Figure 7: Shows two upstream and two downstream flows
In this case the throughput ratio increases because ACK of upstream flow is cluttered at the buffer
of ACK thus the packets of downstream flow experience significant timeout due to packet drop at
the access point buffer thus increasing unfairness.
Case 3 : One upstream and one downstream flow
Figure 8: Shows one upstream and downstream flow
In this case the behaviour depends upon the buffer size B at Access Point and the TCP receiver
window size(w). When this window size is large enough the loss of ACK packet has no influence
on the window size of the sender. This is due to the cumulative acknowledgement nature of TCP
in which the next ACK packet will have the appropriate sequence number thereby compensating
the loss of previous ACK. The upstream window size increases until it reaches w and remains the
same throughout the period of connection. The downstream TCP window size depends upon B
and w. Generally TCP reacts to loss of data packet by halving its window size. Thus if the buffer
size at AP is larger than twice the receiver window size al packets will have space in the buffer
and packets will not be dropped. Fair allocation of bandwidth to the 3 stations i.e. sender, receiver
and access Point by the wireless MAC results in fair allocation of bandwidth to both the TCP
flows.
8. PERFORMANCE OF A* ALGORITHM WITH DCF
The network scenario we deal with is the one having equal number of senders(upload clients) and
receivers(download clients) connected to data source or server through the access point. In this
network, when evaluating the performance of the above stated A* algorithm at access point in the
presence of DCF, the system becomes unstable with more than two uploads.
Figure 9: Shows RTT (in ms) vs number of uploads with A* algorithm when DCF mechanism is applied.
This is due to the combined delay caused by DCF back off interval and packet loss due to eBDP.
SENDER AP RECEIVER
SENDER 1
SENDER 2
RECEIVER 1
RECEIVER 2
AP

Computer
Figure 10: Shows the simulated values of Throughput(Mb/s) vs Time(ms) for Uploads and Downloads in
9. EXPLICIT WINDOW ADAPTATION
In this scheme, the advertised receiver window field present in the ACK packets from the data
source are used. This field is used to indicate the amount of space available at the receiver.
Lowering the value of this field in the access point can help in notifying the upload client that
should reduce the amount of data it sends to the receiver.
9.1 Functioning of EWA
The EWA works as follows. An explicit feedback value is sent to the upload client. This feedback
is present in the advertised receiver window field of the ACK packets.
function of the amount of free space in the access point buffer.
At the access point, if the value in the current advertised receiver window field value W(t) is
greater than the calculated feedback value f(B
the feedback value. The amount of free space in buffer B
B indicates the total buffer size and Q(t) the buffer occupancy at time t. Thus the final feedback
value Wr‘(t) is calculated as a function of the amount of free buffer space at the access point
f(Be(t)) is Wr
’
(t)=min(Wr(t), max(f (B
10. VARIATION OF EWA
Once the modified advertised window field is detected by the upload client, it increases its
queuing time to increase the number of opportunities for the downstream traffic. When the
queuing time is increased, the downstream ACKs get opportunities to retur
hence their clutter at the access point is reduced. This increases the transmission opportunities for
download data thus enhancing fairness.
Upload Traffic
Download Traffic
above stated network scenario.
DAPTATION
receiver window field present in the ACK packets from the data
Lowering the value of this field in the access point can help in notifying the upload client that
should reduce the amount of data it sends to the receiver.
is present in the advertised receiver window field of the ACK packets. This feedback is given as a
function of the amount of free space in the access point buffer.
greater than the calculated feedback value f(Be(t)), the value in the advertised field is reduced to
the feedback value. The amount of free space in buffer Be(t) is calculated as Be(t)=B
ted as a function of the amount of free buffer space at the access point
(t), max(f (Be(t)), MSS)), where MSS is Maximum Segment Size.
EWA TO CREATE FAIRNESS
queuing time is increased, the downstream ACKs get opportunities to return to upload clients and
download data thus enhancing fairness.
363
receiver window field present in the ACK packets from the data
Lowering the value of this field in the access point can help in notifying the upload client that it
This feedback is given as a
the advertised field is reduced to
(t)=B-Q(t) where
ted as a function of the amount of free buffer space at the access point i.e.
(t)), MSS)), where MSS is Maximum Segment Size.
n to upload clients and

Figure 11: shows the simulated values of Throughput(Mb/s) vs Time(ms) for Uploads and
Downloads after EWA variation is implemented in the above n
11.CONCLUSION
Thus by increasing the queuing time in explicit feedback scheme, fairness among the upstream
and downstream flows is achieved when using dynamic buffers. This techni
perfect utilization of the channels bandwidth
be applied only to those routers that can manipulate the transport layer.
REFERENCES
[1] Tianji Li, Douglas Leith and David Malone in “Buffer Sizing for 802.11 Based Networks ”, February
2011 .
[2] Lampros Kalampoukas, Anujan Varm and K. K. Ramakrishnan in “Explicit Window Adaptation: A
Method to Enhance TCP Performance”, June 2002.
[3] Saar Pilosof, Ramachandran Ramjee, Danny Raz, Yuval Shavitt and Prasun Sinha in “Understanding
TCP Fainess Over Wireless Lan’’.
[4] D. J. Leith, P. Clifford, D. Malone, and A. Ng in “TCP Fairness in 802.11e WLANs”, November 2005.
[5] Tianji Li and Douglas J. Leith in “Buffer Sizing for TCP Flows in 802.11e WLANs”, IEEE
communications letters, vol. 12, no. 3, Ma
Upload Traffic
Download Traffic
the simulated values of Throughput(Mb/s) vs Time(ms) for Uploads and
Downloads after EWA variation is implemented in the above network scenario.
and downstream flows is achieved when using dynamic buffers. This technique also leads to
ation of the channels bandwidth. A major disadvantage in this technique is that it can
be applied only to those routers that can manipulate the transport layer.
Method to Enhance TCP Performance”, June 2002.
Saar Pilosof, Ramachandran Ramjee, Danny Raz, Yuval Shavitt and Prasun Sinha in “Understanding
communications letters, vol. 12, no. 3, March 2008.
the simulated values of Throughput(Mb/s) vs Time(ms) for Uploads and
etwork scenario.
que also leads to
. A major disadvantage in this technique is that it can
Saar Pilosof, Ramachandran Ramjee, Danny Raz, Yuval Shavitt and Prasun Sinha in “Understanding

AUTHORS
Bhuvana Lalitha N is a graduate in B.Tech Information Technology from Easwari
Engineering College, Chennai, India and is joining Tata Consultancy Services in
August 2012 . Her research interests are in the field of Wireless Networks and
Robotics. Email: bhuvana.n21@gmail.com
Divya A, a graduate in B.Tech Information Technology from Easwari Engineering
College, Chennai, India, campus-selected in Tata Consultancy Services in 2012 . Her
research interests are in the field of Telecommunication and Networks.
Email: divi_cool8@yahoo.com
Divya P D, is a graduate in B.Tech Information Technology from Easwari
Engineering College, Chennai, India. She will be joining Tata Consultancy Services
shortly . Her research interests are in the field of Networks and Router security.
Email: div.pd91@gmail.com

ENHANCEMENT OF TCP FAIRNESS IN IEEE 802.11 NETWORKS

More Related Content

What's hot (19)

Similar to ENHANCEMENT OF TCP FAIRNESS IN IEEE 802.11 NETWORKS (20)

More from cscpconf (20)

Recently uploaded (20)

ENHANCEMENT OF TCP FAIRNESS IN IEEE 802.11 NETWORKS