CONGESTION CONTROL WITH DYNAMIC THRESHOLD
ADAPTATION AND CROSS-LAYER RESPONSE FOR TCP
OVER IEEE 80.11 WIRELESS NETWORKS
Rung-Shiang Cheng
1
, Der-Jiunn Deng
2
and Han-Chieh Chao
3
1
Department of Computer and Communication, Kun Shan University, Yung Kang, Taiwan
2
Department of Computer Science and Information Engineering, National Changhua University of Education
Changhua City, Taiwan
3
Department of Electronic Engineering, National Ilan University, Ilan City, Taiwan
Keywords: TCP, Congestion control, IEEE 802.11, Cross-layer response.
Abstract: Wireless technologies provide mobile access and enable rapid and cost-effective network deployment. But a
wireless link is generally accompanied by high interference, transmission errors and a varying latency. The
erratic packet losses usually lead to a curbing of the flow of segments on the TCP connection, and thus limit
TCP performance. This paper presents a threshold control mechanism with cross-layer response approach
for improving TCP Vegas performance in IEEE 802.11 wireless networks. By making slight modifications
to the legacy IEEE 802.11 MAC and TCP, the numerical results reveal that the proposed scheme provides a
significant improvement in TCP performance under IEEE 802.11 wireless environments.
1 INTRODUCTION
The proliferation of mobile computing devices in
recent years has led to the extensive deployment of
wireless communication networks. More and more
devices (such as PCs, Laptops, Smart phones,
Tablets) are equipped with network interface cards
that integrate wireless technologies (such as
3G/UMTS/GPRS and Wi-Fi) to connect to the
networks. New standards for wireless
communications (IEEE Standard 802.11, 1999)
(IEEE Standard 802.16-2004, 2004) have greatly
improved data transmission rate and prompted the
development of high-speed wireless communication
services capable of supporting the anticipated
introduction of a wide range of new applications
over the coming years.
Recently, TCP and IEEE 802.11 have become
the most commonly applied transport protocol for
Internet applications and the standard of choice for
wireless local area networks, respectively. In IEEE
802.11, wireless communication applications are
commonly constrained by scalability problems.
which fail to achieve a reasonable throughput as the
number of hosts increases since CSMA/CA access
schemes have extremely low spatial-reuse
efficiency. As a result, the link layer reliability
deteriorates rapidly as the scale of the network
increases, and thus the number of link layer
contentions increases significantly.
TCP Reno (Paxson, 1999) is currently the most
widely applied transport layer protocol for Internet
congestion control. Although Reno-based TCP (e.g.
NewReno, SACK) (Floyd, 1999) (Mathis, 1996) is
the most commonly used TCP, however, its
aggressive behavior causes a severe oscillation of
the congestion window size, and the resulting
periodic congestion leads to a limited throughput.
Hence, several bandwidth estimation schemes for
TCP have been proposed (Jain, 2002). Some TCP
bandwidth estimation schemes utilize the end-to-end
delay information derived from the measured packet
round-trip time (RTT). The delay-based approach,
originally described by Jain (Jain, 1989), is best
represented by TCP Vegas (Brakmo, 1995).
Unlike Reno-based TCP, TCP Vegas implements
a more precise congestion avoidance mechanism
based on packet delay rather than the packet loss.
Although TCP Vegas yields both a higher
throughput and a lower packet-loss ratio than TCP
Reno does, however, TCP Vegas fails to obtain a
fair share of the bandwidth when competing with
TCP Reno in heterogeneous networks. Therefore, in
(Cheng, 2010), we proposed an adaptive TCP
95
Cheng R., Deng D. and Chao H..
CONGESTION CONTROL WITH DYNAMIC THRESHOLD ADAPTATION AND CROSS-LAYER RESPONSE FOR TCP OVER IEEE 80.11 WIRELESS
NETWORKS.
DOI: 10.5220/0003677100950100
In Proceedings of the International Conference on Wireless Information Networks and Systems (WINSYS-2011), pages 95-100
ISBN: 978-989-8425-73-7
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
congestion control scheme which dynamically
adjusts the threshold parameters being used to define
the congestion window size in such a way as to
improve the buffer occupancy of TCP Vegas when
competing with TCP Reno.
Since the performance of data transfer not only
relies on the support of the end-to-end transport
layer, but also on the quality of the hop-by-hop
communication link, the present study attempts to
enhance the legacy IEEE 802.11 MAC and TCP
protocols such that TCP is rendered capable of
differentiating between corruption and congestion
losses by using information received from the MAC
layer. In this paper, the current study proposes a
collaborative cross-layer approach based on the
threshold adjustment mechanism and negative-ACK
scheme for enhancing the performance of TCP
Vegas in wireless networks.
The remainder of this paper is organized as
follows. Section 2 briefly describes the IEEE 802.11
media access control scheme and the TCP protocol.
Section 3 describes the implementation of the cross-
layer TCP Vegas and IEEE 802.11 MAC protocol.
Section 4 compares the performance of the cross-
layer scheme with that of TCP Vegas by performing
a series of numerical simulations. Finally, Section 5
presents some brief conclusions.
2 RELATED WORKS
2.1 Media Access in IEEE 802.11 MAC
The IEEE 802.11 standard defines two types of
service, namely a contention-free polling-based
point coordination function (PCF) and a contention-
based distributed coordination function (DCF). In
infrastructure-based networks, DCF can operate
either alone or in conjunction with PCF. However,
in ad hoc networks, DCF operates alone. DCF is the
basic access method for the 802.11 standard and is
based on the conventional carrier sense multiple
access with collision avoidance (CSMA/CA)
scheme. DCF comprises both a basic access method
and an optional channel access method based on
RTS/CTS exchanges.
In 802.11, priority access to the wireless medium
is controlled by applying an inter-frame space (IFS)
time between the frame transmissions. To prevent
collisions, the transmitter is obliged to wait for the
availability of the channel for a specified interval of
time, designated as the distributed inter-frame space
(DIFS) before sending a frame. If the medium is
currently busy or becomes busy during this interval,
the transmitter defers the frame transmission until it
detects a DIFS.
Due to the half-duplex nature of wireless
interfaces, stations in the network are unable to
detect a collision simply by listening to their own
transmissions. Therefore, an immediate positive
acknowledgment technique is employed to confirm
the successful reception of a frame. Specifically,
having received a frame, the receiver waits for a
short inter-frame space (SIFS) and then transmits a
positive MAC acknowledgment to the transmitter
confirming that the frame has been correctly
received. The SIFS is deliberately assigned a shorter
interval than the DIFS in order to assign the
receiving station a higher priority than any other
stations waiting for making a transmission. The
ACK is only transmitted when the frame is received
correctly, and hence if the transmitter does not
receive an ACK, it assumes that the data frame must
have been lost and therefore schedules a
retransmission. Figure 1(a) illustrates the basic
operations involved in 802.11 DCF.
To alleviate the hidden-station problem (Kim,
1997), 802.11 DCF also provides an optional
channel access method using a virtual carrier sensing
mechanism based on two special control frames,
namely request-to-send (RTS) and clear-to-send
(CTS). As shown in Figure 1(b), before transmitting
a frame, the transmitter transmits an RTS frame
asking the receiver if the medium in its vicinity is
free. Once the receiver receives this RTS frame and
assumes that no interfering transmission is present, it
waits for the specified SIFS interval and then sends a
CTS frame to the transmitter. Both transmitter and
receiver neighbors overhear these frames and
consider the medium reserved for the transmission
duration.
(a) Basic access mechanism.
(b) Virtual carrier sensing mechanism.
Figure 1: Media access mechanisms in IEEE 802.11 DCF.
The IEEE 802.11 MAC layer provides a reliable
communication link by handling packet delivery
using the MAC positive ACK and retransmission
WINSYS 2011 - International Conference on Wireless Information Networks and Systems
96
mechanisms (i.e. an automatic repeat request,
(ARQ)). However, if several transmission attempts
of a packet (e.g. 7 times for the basic access
mechanism and 4 times for the virtual carrier
sensing mechanism) by a sender in a wireless
network fail, the sender simply assumes that the
receiver is no longer within its transmission range
and gives up its transmission attempt.
As described above, the stop-and-wait schemes
such as the IEEE 802.11 ARQ scheme are highly
inefficient due to the idle time spent by the sender
waiting for the receiver's positive ACK message to
arrive. When the end-to-end path comprises multiple
wireless links, the accumulated backoff delays and
losses further reduce the network throughput, and
thus increasing the underutilization of the wireless
links. Hence, it is desirable to enhance the
interaction between the MAC layer and the transport
layer in order to achieve the level of transmission
throughput demanded by communication protocols
such as TCP.
2.2 Congestion Control in TCP Vegas
TCP Vegas (Brakmo, 1995) adopts the measured
RTT to calculate the number of data packets which
the sender can inject into the network without
incurring packet loss. While doing this, Vegas
attempts to maintain the number of the packets
queued in the router buffer within two limits,
denoted as α and β, respectively. If the measured
RTT increases, Vegas recognizes that the network is
becoming congested and throttles the congestion
window down. Conversely, if the RTT reduces,
Vegas recognizes the availability of more bandwidth
and therefore increases the congestion window size
accordingly. Each time a new ACK is received, the
sender updates the congestion window size in
accordance with the difference (δ) between the
expected throughput and the actual throughput.
According to the definition in (Chengrs, 2010), δ is
represented as the number of the packets queued in
the buffer of end-to-end path and can be expressed
as
()
minRTT
TmeasuredRT
W
minRTT
W
minRTT
×
=
×= t throughpuActualt throughpuExpected
δ
(1)
Here, the expected throughput represents the
available bandwidth for the current TCP connection
in the absence of network congestion, while the
actual throughput represents the bandwidth currently
used by the TCP connection. Finally, minRTT repre-
sents the minimum value of the measured RTT.
Vegas utilizes two threshold parameters α and β
to control the maximum difference between the
expected and actual throughputs. The congestion
window size, W, is increased by one packet if δ < α
and is decreased by one packet if α > β (the default
values of α and β are 1 and 3, respectively). In other
words, having calculated
δ
, the sender updates W in
accordance with
>
<+
=
otherwise,
if,1
if,1
W
W
W
W
βδ
αδ
(2)
To detect and avoid congestion during the slow-start
phase, Vegas doubles the window size every two
RTT intervals rather than every RTT interval as it
does in Reno. Furthermore, when receiving a
duplicate ACK, the sender checks whether the
difference between the current time and the sending
time plus the minRTT of the relevant packet is
greater than the timeout value. If it is, Vegas
retransmits the packet immediately without waiting
for the arrival of three duplicate ACKs. This
modification resolves the problem that the sender
never actually receives three duplicate ACKs and
must therefore rely on the occurrence of the coarse-
grain timeout and therefore yields a significant
improvement in the throughput.
3 DESCRIPTION OF PROPOSED
METHOD
The IEEE 802.11 link layer protocol supports
reliable data to transfer locally while the TCP
transport layer protocol supports reliable data to
transfer end-to-end. This section proposes a
collaborative approach for TCP Vegas enhancement.
3.1 Extension to IEEE 802.11 MAC
In standard IEEE 802.11 DCF schemes, whenever a
node fails to transmit a frame (e.g. channel
collisions), it retransmits that frame and then
increases the value of the retry limit parameter by
one. However, if the retry transmission exceeds a
specified retry threshold, the IEEE 802.11 MAC
protocol reports a link breakage and then discards
the transmitting frame. To reduce the effects of link
layer packet corrupt on the performance of TCP, this
study extends the IEEE 802.11 DCF. Whenever a
frame transmission failure occurs, a negative-ACK
option is triggered by modified link layer protocol in
accordance with the following function:
CONGESTION CONTROL WITH DYNAMIC THRESHOLD ADAPTATION AND CROSS-LAYER RESPONSE FOR
TCP OVER IEEE 80.11 WIRELESS NETWORKS
97
If the number of retry limit parameter exceeds a
specified threshold {
If receiving of TCP ACK within the same flow
{
Forward the negative-ACK option piggybacks
using the reverse TCP ACK along the end-to-end
path. }
}
Else {
Increase the contention window and then launch
the backoff procedure.
}
In the extended protocol, if a TCP data frame is
discarded after several retransmission attempts, the
modified link layer protocol triggers a negative-
ACK as the sequence number associates with the
dropped packet and then piggybacks this option by a
reverse TCP ACK in order to avoid increasing
contention in the link layer. This cross-layer support
from the link layer protocol ensures that the
transport layer TCP protocol is aware of the
transmission error in the link layer and can then
react to this error in accordance with the wireless
corruption information.
The extended link layer protocol provides
additional information for natively wired TCP
protocols to the wireless environment and for
improving their performance. A major advantage of
the proposed scheme is that it eliminates the
requirement of forward nodes to cache
unacknowledged packets for every TCP connection
passing through it and is required to be fully
supported at all of the wireless stations in the
wireless network.
3.2 Modification to Vegas Congestion
Control
To estimate the buffer currently available in each
round-trip time interval, the expected bandwidth for
TCP at the bottleneck link in the network is first
derived. The queue length at the bottleneck is then
approximately measured in each round-trip time
interval from the inter-arrival times of the returned
ACK messages. The available buffer size along the
path between the sender and the receiver is then
estimated from the derived value of the expected
bandwidth and the measured value of the queue
length. Finally, the TCP Vegas threshold parameters
α
and
β
are adjusted in accordance with the
estimated buffer length, thereby, changing the size
of the congestion window indirectly and
consequently improving the network utilization. The
details of the proposed Vegas transmission control
scheme are described in the following.
Let maxRTT be the maximum measured RTT,
and λ denote the bottleneck link bandwidth as
calculated by the packet-pair scheme (i.e. the
number of the acknowledged packets divided by the
average ACK inter-arrival time) (Keshav, 1991). Let
Q
m
represent the measured buffer size in last RTT.
According to the information obtained from last
RTT, Q
m
can then be calculated as:
()
1
m
WW W
Q
minRTT maxRTT maxRTT
minRTT
ρρλ


=⋅ +




×
(3)
where
ρ
is a smoothing factor, and Q
m
is obtained by
the exponentially weighted moving average (EWMA)
method.
Let Q
t
denote the average observed buffer size
which can be held in the end-to-end path. To prevent
the estimated queue length from wild fluctuation, Q
t
is configured by using the EWMA as:
Q
t
=
ν
· Q
t-1
+
()
ν
1 ·
m
Q
(4)
where the parameter,
ν
indicates the ratio of TCP
sender transmission time to its measured round-trip
delay time. The value of
ν
varies with the bandwidth
utilization of the sender in the previous round-trip.
According to Eq. (4), the threshold values α and
β are then adjusted as follows:
2-1 where
timeoutgrainedcoarseif),default(1
losspacket if,2
durationpreset
a beyond status stable in staying continueif,2
status stable in staying coutinueif,
t
t
Q
W
Q
=
α
α
α
α
(5)
1.5. is of luedefault va theand ,3 where
timeoutgrainedcoarseif),default(3
mQ
m
t
=
β
α
β
(6)
When receive an ACK, the modified Vegas
regulates its transmission rate in accordance with the
following function:
If receiving a new ACK {
Adjust the threshold parameters α and β according to
Eqs. (5) and (6) and then regulates the congestion
window size based on the following function that
mentioned in Eqs. (1) and (2):
>
<+
=
otherwise,
if,1
if,1
W
W
W
W
βδ
αδ
}
Else {
If receiving a triple-duplicate ACK, and the
WINSYS 2011 - International Conference on Wireless Information Networks and Systems
98
next expected packet is not a "corrupted
packet" {
Retransmit the next expected packet and set W
t
= W/2 and W = W
t
. Then modified Vegas enter
the regular TCP Fast recovery phase. }
}
If receiving the ACK which contains a negative-
ACK option {
Recording sequence number of the "corrupted
packet" obtained from negative-ACK, and then
retransmitting the corrupted packet.
}
In the proposed approach, the negative-ACK number
of the corrupted packet is tracked using a circular
queue. By inspecting the historical negative-ACK
information, the modified TCP can clearly identify
corruption losses in the wireless links and therefore
determine whether it is necessary to invoke the
regular congestion control mechanisms. Hence, if
retransmitted packets are lost repeatedly as a result
of transmission errors, the original packets can be
retransmitted immediately upon receipt of an ACK
with a negative-ACK option without waiting for
timeout expiry or for the "round-trip" time to elapse
as in conventional TCP error recovery schemes.
4 NUMERICAL RESULTS
In the simulations, the ns-2 (Network Simulator, NS-
2) 802.11 MAC layer and transport layer modules
were both extended to model the proposed cross-
layer functionality. The parameter values used in the
simulations are summarized in Table 1. The default
sizes of RTS, CTS and ACK packets are assigned to
20 bytes, 14 bytes and 14 bytes, respectively.
Furthermore, the two-ray ground reflection model
(Rappaport, 1960) was used to simulate signal
propagation in the wireless network.
Table 1: Default parameters for MAC and physical layers.
Parameter 802.11g
SLOT 9 μsec
SIFS 10 μsec
DIFS 28 μsec
PHY
hdr
192 bits
CW
min
32
CW
max
1024
In the following simulations, packet errors are
introduced into the wireless link (indicated by dotted
line) using the two-state Markov error model
(Gilbert, 1960) (see Figure 2). The packet corruption
rates P
G
and P
B
are set as 0.0005 and 0.001,
respectively, and P
GG
, P
GB
, P
BB
and P
BG
are set to
0.9, 0.1, 0.85 and 0.15, respectively. The rate at
which errors occur in the GE model depends on
channel conditions.
Figure 2: Two-state Markov error model.
Figure 3 gives the simulation topology when two
TCP flavors (Reno and Vegas) coexist.
Figure 3: Simulation model with link-level errors.
Table 2 compares the proposed Vegas
modification with the original Vegas. As shown in
Table 2, packets transmitted on wireless channels
are frequently affected by burst errors which cause
the MAC layer to discard the frame, resulting in the
loss of the corresponding packet in the TCP
protocol. In this case, although Reno still has a better
performance, it can be seen that the variation
between mechanisms is decreased. Transmission
rate of Reno is not able to increase stably because of
bit error and provide Vegas more opportunities to
obtain more bandwidth. From results shown in Table
2, it can be seen that the proposed Vegas
modification decreases the variation between the
two mechanisms and achieves better bandwidth
utilization.
Table 2: Interaction of TCP in heterogeneous
environments (with link-level bit error).
TCP flavors Average goodput
Reno:Vegas 23.59 Mbps:16.53 Mbps
Reno:Modieied Vegas 19.04 Mbps:19.76 Mbps
To further examine the performance of the
proposed scheme under a more realistic and
complicated network environment, a series of
simulations is conducted on randomly generated ad
hoc network topologies. In these simulations, the
CONGESTION CONTROL WITH DYNAMIC THRESHOLD ADAPTATION AND CROSS-LAYER RESPONSE FOR
TCP OVER IEEE 80.11 WIRELESS NETWORKS
99
network region is 100 m × 100 m with 20 wireless
stations, and TX
range
and PCS
range
are set to 40 m and
85 m, respectively. As illustrated in Figure 4, at the
beginning of each simulation, wireless stations are
randomly placed within this region, and a random ad
hoc topology is constructed. The number of short-
lived TCP connections varies from two to nine.
Moreover, the TCP source and the TCP destination
of each connection are randomly selected.
Figure 4: Illustrative random topology.
Table 3 presents the TCP goodputs of the
proposed mechanism and other TCP flavors when
the path lengths are 2-hop and 3-hop, respectively.
As shown in Table 3, the TCP goodput is inversely
proportional to the number of connections. This
phenomenon is caused by the competition for the
bandwidth among TCP connections and the carrier
sense dependencies over wireless stations.
Comparing the results in Table 3, it is obvious that
the proposed Modified Vegas achieves a better
performance than those of other TCP schemes.
Table 3: TCP average goodput comparison (Only the TCP
goodput with established connection is calculated).
Number of
connection
Reno Vegas Modified Vegas
2 11.58Mbps 11.51Mbps 14.52Mbps
3 9.76 Mbps 10.32Mbps 12.51Mbps
4 8.75 Mbps 8.74 Mbps 10.82Mbps
5 7.24 Mbps 8.20 Mbps 9.32 Mbps
6 6.53 Mbps 7.11 Mbps 8.90 Mbps
7 6.50 Mbps 7.04 Mbps 8.37 Mbps
5 CONCLUSIONS
This paper has proposed a cross-layer approach
designated to enhance the performance of TCP
Vegas in wireless networks. In the proposed
approach, a cross-layer mechanism is employed in
both the transport layer and the MAC layer to
provide explicit corruption loss information. The
results confirm that the proposed scheme has a
number of key advantages compared to the
conventional TCP, including a more efficient
treatment on frequent transmission losses, a faster
reaction to corruption losses, and the ability to
distinguish between congestion errors and
transmission errors and to take appropriate remedial
action which is particularly advantageous for the
deployment in heterogeneous wireless environments.
REFERENCES
IEEE Standard 802.11, Wireless LAN Medium Access
Control (MAC) and Physical Layer (PHY)
Specifications
, 1999.
IEEE Standard 802.16-2004,
Part 16: Air Interface for
Fixed Broadband Wireless Access System
, 2004.
Paxson, V., Allman, M., Stevens, W.,
TCP Congestion
Control
, RFC 2581, 1999.
Floyd, S., Henderson, T.,
The NewReno Modification to
TCP's Fast Recovery Algorithm
, RFC 2582, April
1999.
M. Mathis, J. Mahdavi, S. Floyd, A. Romanow, TCP
Selective Acknowledgement Options, RFC 2018,
April 1996.
M. Jain, C. Dovrolis, End-to-End Available Bandwidth:
Measurement Methodology, Dynamics, and Relation
with TCP Throughput, IEEE/ACM Transactions on
Networking, Vol. 11, No. 4, pp. 537
549, 2002.
R. Jain, A Delay-based Approach for Congestion
Avoidance in Interconnected Heterogeneous Computer
Networks, Computer Communication Review, Vol. 19,
No. 5, pp. 5671, 1989.
L. Brakmo, L. Peterson, TCP Vegas: End to End
Congestion Avoidance on a Global Internet, IEEE
Journal on Selected Areas in Communications, Vol. 13,
No. 8, pp. 1465
1480, 1995.
R. S. Cheng, M. Y. Shih, C. C. Yang, Performance
Evaluation of Threshold-based Control Mechanism for
Vegas TCP in Heterogeneous Cloud Networks,
International Journal of Internet Protocol Technology,
Vol. 5, No. 4, pp. 202-209, March 2010.
J. Kim, B. Crow, I. Widjaja, P. Sakai, IEEE 802.11
Wireless Local Area Networks, IEEE
Communications, Vol. 35, No. 9, pp. 116
126, 1997.
S. Keshav, A Control-Theoretic Approach to Flow
Control, ACM SIGCOMM Computer Communication
Review, Vol. 21, No. 4, pp. 3-15, 1991.
Network Simulator, NS-2, Available:
http://www.isi.edu/nsnam/ns/.
T. S. Rappaport and T. Rappaport, Wireless
Communications: Principles and Practice, 2nd Edition,
Prentice Hall, 2001.E. N. Gilbert, Capacity of a Burst-
noise Channel, Bell System Technical, Vol. 69, pp.
1253
1265, 1960.
E. N. Gilbert, Capacity of a Burst-noise Channel, Bell
System Technical, Vol. 69, pp. 1253
1265, 1960.
WINSYS 2011 - International Conference on Wireless Information Networks and Systems
100