Content uploaded by Lemin Li
Author content
All content in this area was uploaded by Lemin Li on Jan 16, 2013
Content may be subject to copyright.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 4, JULY 2004 1235
A New Collision Resolution Mechanism to Enhance
the Performance of IEEE 802.11 DCF
Chonggang Wang, Bo Li, Senior Member, IEEE, and Lemin Li
Abstract—The medium-access control (MAC) protocol is one
of the key components in wireless local area networks (WLANs).
The main features of a MAC protocol are high throughput, good
fairness, energy efficiency, and support priority guarantees,
especially under distributed contention-based environment. Based
on the current standardized IEEE 802.11 distributed coordination
function (DCF) protocol, this paper proposes a new efficient col-
lision resolution mechanism, called GDCF. Our main motivation
is based on the observation that 802.11 DCF decreases the con-
tention window to the initial value after each success transmission,
which essentially assumes that each successful transmission is an
indication that the system is under low traffic loading. GDCF
takes a more conservative measure by halving the contention
window size after
consecutive successful transmissions. This
“gentle” decrease can reduce the collision probability, especially
when the number of competing nodes is large. We compute the
optimal value for
and the numerical results from both analysis
and simulation demonstrate that GDCF significantly improve the
performance of 802.11 DCF, including throughput, fairness, and
energy efficiency. In addition, GDCF is flexible for supporting
priority access by selecting different values of
for different traffic
types and is very easy to implement it, as it does not requires any
changes in control message structure and access procedures in
DCF.
Index Terms—IEEE 802.11 DCF, wireless local area network
(WLAN).
I. INTRODUCTION
R
ECENTLY, we have witnessed a rapid development and
deployment of wireless local area networks (WLANs),
which in return has fueled the development in the standardiza-
tion organization, such as the IEEE 802.11 working group, to
improve its performance. One of the key components in WLAN
is a medium-access control (MAC) protocol that primarily
determines its performance. MAC protocols are commonly
used in multiple-access environments, where multiple nodes
compete for certain shared resources. The main functionality
Manuscript received November 9, 2003; revised January 9, 2004. B. Li’s re-
search was support in part by grants from the Research Grant Council under
contracts HKUST6196/02E and HKUST6402/03E, an National Science Funds
Council/RGC joint grant under contract N_HKUST605/02, and a grant from
Microsoft Research under contract MCCL02/03.EG01.
C. Wang is with the the Special Research Centre for Optical Internet & Wire-
less Information Networks (ICOIWIN), ChongQing University of Posts and
Telecommunications (CQUPT), Chongqing 400065, P. R. China, on leave from
the Department of Computer Science and Computer Engineering, University of
Arkansas, Fayetteville, AR 72701 USA (e-mail: cgwang@cs.ust.hk).
B. Li is with the Department of Computer Science, The Hong Kong Univer-
sity of Science and Technology, Hong Kong, P.R. China (e-mail: bli@cs.ust.hk).
L. Li is with the University of Electronic Science and Technology of China,
Chengdu 400065, P. R. China (e-mail: lml@uestc.edu.cn).
Digital Object Identifier 10.1109/TVT.2004.830951
of MAC protocols is to arbitrate access for the shared transmis-
sion medium [1]. The performance metrics of interest include
throughput, fairness, packet transmission delay, stability, and
also priority in an environment supporting multiservices. In
addition, in a WLAN, the energy efficiency is also a major
performance index of interest.
In IEEE 802.11 standard [2], channel access is controlled by
the use of interframe space (IFS) time between the frame trans-
missions. Three IFS intervals that have been specified by 802.11
standards include short IFS (SIFS), point coordination function
IFS (PIFS), and distributed coordination function (DCF)-IFS
(DIFS). The SIFS is the smallest and the DIFS is the largest.
There are two access mechanisms including point coordina-
tion function (PCF) and DCF. PCF is a centralized MAC algo-
rithm used to provide contention-free service, while DCF uses a
contention-based algorithm to provide access to all traffic. PCF
is built on top of DCF and regulates transmission through a cen-
tralized decision maker or point coordinator, which makes use of
PIFS when issuing polls. Because PIFS is smaller than DIFS, the
point coordinator can seize the medium and lock out all asyn-
chronous traffic (which uses DIFS to access channel) while it
issues polls and receives responses. This paper focuses on DCF
and we will give a brief introduction later.
In 802.11 DCF, a node starts its transmission if the medium
is sensed to be idle for an interval larger than the distributed
interframe space (DIFS). If the medium is busy, the node will
defer its transmission until a DIFS is detected and then generate
a random backoff period (backoff timer) before retransmission.
The backoff timer will be decreased as long as the channel is
sensed idle, frozen when the channel is sensed busy, and re-
sumed when the channel is sensed idle again for more than
a DIFS. A node can initiate a transmission when the backoff
timer reaches zero. The backoff timer is uniformly chosen in
the range
CW). CW is known as contention window, which
is an integer with the range determined by the PHY character-
istics CW
and CW . After each unsuccessful transmis-
sion, CW will be doubled until reaching the maximum value
CW
, where equals to (CW . After
each successful transmission, CW will reset to the minimum
value CW
. In 802.11 DCF for the DSSS physical channel,
CW
,CW , and .
802.11 DCF defines two channel-access modes: basic and re-
quest to send/clear to send (RTS/CTS) base access. In basic ac-
cess mode [Fig. 1(a)], the destination node will wait for a SIFS
interval immediately following the successful reception of the
data frame and transmit a positive ACK back to the source node
to indicate that the data packet has been received correctly. If the
source node does not receive an ACK, the data frame is assumed
to be lost and the source node will schedule the retransmission
0018-9545/04$20.00 © 2004 IEEE
1236 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 4, JULY 2004
Fig. 1. IEEE 802.11 MAC mechanism.
with the doubled CW for backoff timer. When the data frame
is being transmitted, all the other nodes hearing the data frame
adjust their network-allocation vector (NAV), which is used for
virtual carrier sense at the MAC layer, correctly based on the
duration field value in the data frame received. This includes
the SIFS and ACK frame transmission time following the data
frame.
In RTS/CTS-based access mode, nodes transmit data utilizing
special short RTS and CTS frames prior to the transmission of an
actual data frame in order to shorten the collided time interval.
As shown in Fig. 1(b), the node that needs to transmit a packet is-
sues a RTS frame. When the destination receives the RTS frame,
it will transmit a CTS frame after the SIFS interval, immediately
following the reception of the RTS frame. The source node is
allowed to transmit its packet if and only if it receives the CTS
correctly. At the same time, all the other nodes will update the
NAVs based on the RTS from the source node and the CTS from
the destination node, which helps to overcome the hidden ter-
minal problem. In fact, the node that is able to receive the CTS
frames correctly can avoid collisions even when it is unable to
sense the data transmissions from the source node. If a collision
occurs with two or more RTS frames, less bandwidth is wasted
as compared to the situation where larger data frames can col-
lide in the basic access mode.
The remainder of this paper is organized as follows. Section II
reviews the existing work and discusses the main features in our
proposed GDCF. Section III introduces the new collision-reso-
lution mechanism called GDCF. Theoretical analysis of GDCF,
including normalized throughput and some other metrics, will
be given in Section IV. In Section V, we present numerical re-
sults of GDCF and compare with that those of the IEEE 802.11
DCF protocol. Section VI concludes this paper.
II. R
ELATED WORK
This paper focuses on the contention-based MAC protocols
used in WLAN, specifically IEEE 802.11 DCF [2]. The
analysis in [3] demonstrated that the throughput and fairness of
802.11 DCF could significantly deteriorate when the number
of nodes increases. Several recent proposals have addressed
this issue [4]–[7]. Furthermore, given the need to support
multimedia applications and to consider the energy efficiency
in mobile devices, there also are protocols to address a priority
scheme in [8] and [9] and the energy efficiency issue in [10].
Cali et al. [4] proposed a dynamic and distributed algorithm,
IEEE 802.11
, which allows each node to estimate the number
of competing nodes and to tune its contention window to the
optimal value at run time. Results from simulations showed that
the throughput of IEEE 802.11
is very close to the theoretical
upper bound. DCF
, proposed in [5], is a new ACK-integrated
mechanism that combines the TCP ACK with MAC level ACK
and obtains the improved throughput. One of the limitations
is its ineffectiveness for other flows, such as UDP. It also
violates the layering principle that leads to the complication
in MAC ACK message structure. Peng [6] proposed a new
measurement-based algorithm to adaptively configure the
optimal value of the initial CW value to improve the throughput
and fairness. However, it also needs to compute current
channel status at run time and adjusts the RTS/CTS message
structure. The fast collision resolution (FCR) is another MAC
protocol proposed in [7], which actively redistributes the
backoff timer for all competing nodes, thus allowing the more
recent successful nodes to use smaller contention window and
allowing other nodes to reduce backoff timer exponentially
when they continuously meets some idle time slots, instead
of reducing backoff timer by 1 after each idle time slots, as
in the original IEEE 802.11 DCF. FCR can resolve collisions
more quickly than 802.11 DCF and obtains higher throughput,
but FCR itself can inversely affect the fairness unless it is
combined with additional fair scheduling mechanism, as shown
in [7]. Residual-energy-based tree splitting (REBS) [10] is an
energy-efficient collision-resolution algorithm that can be used
in the wireless ad hoc networks. REBS differentiates and splits
all the competing nodes according to their residual energy and
assigns the node with the least residual energy to seize the
channel with the highest priority.
We propose a new collision-resolution mechanism called
GDCF, which is a simple variation of 802.11 DCF, yet can
significantly improve throughput and fairness. GDCF enables
the priority support for multimedia application and obtains
better energy efficiency than DCF itself. There are several
unique advantages in the proposed GDCF. Comparingit to
IEEE 802.11
in [4] and self-adapt DCF in [6], GDCF is
simpler in that it does not need to estimate network parameters
such as competing node number in [4] and channel status
in [6], although there is a Kalman filter-based algorithm to
measure the number of competing nodes in [11]. Comparing it
to DCF
in [5], GDCF can support any upper protocols (TCP
WANG et al.: A NEW COLLISION RESOLUTION MECHANISM TO ENHANCE THE PERFORMANCE OF IEEE 802.11 DCF 1237
Fig. 2. Collision-resolution stage evolution in GDCF and 802.11 DCF.
or UDP) and does not need to change the RTS/CTS message
structure. Comparing it to the FCR algorithm in [7], GDCF
achieves better fairness and simplicity and can easily support
priority or quality-of-service (QoS) differentiation effectively.
GDCF maintains excellent compatibility with the original IEEE
802.11 DCF. In summary, the proposed GDCF achieves better
throughput, fairness, and energy efficiency and enables priority
support. In addition, it does not need to estimate the competing
node and channel status; thus, it is simple for implementation.
III. P
ROPOSED
GDCF ALGORITHM
From the discussions in the Section II, we can see that
802.11 DCF resolves collision through CW and backoff stage
[Fig. 2(a)]. In the initial backoff stage (stage 0), the value of
CW has the minimal value CW
. After each transmission
collision, the backoff stage will be increased by 1 and the CW
will be doubled until it reaches the maximum, CW
. After
each successful transmission, the backoff stage will resume to
initial stage 0 and the CW will be reset to CW
, regardless
of network conditions such as the number of competing nodes.
In this method, we refer to heavy decrease, which tends to
work well when there are only a few competing nodes. When
the number of competing nodes increases, it will be shown to
be ineffective, since new collisions can potentially occur and
cause significant performance degradation.
For example, assuming that the current backoff stage is
with contention window CW and that there is
a successful transmission, the next backoff stage will be stage 0
with CW
, according to 802.11 DCF specifications for
DSSS PHY [1]. But if the number of current competing nodes
is large enough,
, the new collision will likely occur at
the backoff stage 0. The main argument is that since the current
backoff stage is
, some collisions must have occurred recently.
Now if the number of current competing nodes is larger than
or close to CW
and if the backoff stage is reset to 0 after a
successful transmission, there is a high probability that some
new collision(s) will happen. Certainly, the number of current
competing nodes may be smaller than CW
if there are sev-
eral consecutive successful transmissions at the backoff stage
.
Under this case, we can effectively begin to decrease the CW.
This is the primary principle used in GDCF.
GDCF attempts to avoid useless collisions through the
“gentle” decrease of contention window, referred as gentle
DCF or GDCF. The collision-resolution stage evolution in
GDCF is presented in Fig. 2(b). The difference between GDCF
and DCF is that GDCF will halve CW value if there are
consecutive successful transmissions. On the contrary, DCF
will reset CW once there is a successful transmission or the
retry count overruns the threshold
, so GDCF needs to
maintain a counter for recording the number of continuous
successful transmissions up to now. This counter will reset to
zero after each collision, because what it records is the number
of continuous successful transmissions, not the number of total
successful transmissions. According to the channel status, the
detailed collision resolution process in GDCF is as follows.
•
Collision: Similar to the operations in DCF, GDCF will
double the contention window and select a backoff timer
value uniformly from
CW]. But GDCF also needs to
reset the counter for recording the number of consecutive
successful transmissions.
• Successful transmission: If there are
consecutive suc-
cessful transmissions, GDCF will halve the CW and select
a backoff timer value uniformly from
CW]. Then, the
counter for recording the number of continuous successful
transmissions is reset to zero. Otherwise, GDCF increases
counter for the number of consecutive successful trans-
mission and keeps the contention window unchanged.
• Idle: If the channel is idle, GDCF also reduces the backoff
timer by 1, the same as in DCF.
In another word, CW in GDCF is gently and gradually de-
creased after consecutive successful transmissions. If there are
a few competing nodes, many consecutive successful transmis-
sions can appear and the backoff stage gradually goes down
to the initial stage 0. If there are many competing nodes, the
probability that the backoff stage will be down to initial stage 0
are very small and the backoff stage will oscillate between two
large stages
and with a high probability. This behavior
brings two advantages. First, it will decrease the collision prob-
ability and improve the system throughput. Second, it will ob-
tain better fairness because GDCF maintains all the nodes in
the same stage (with the same CW) even if after several con-
secutive successful transmissions
, especially under large
node number. However, nodes in DCF will stay in a different
stage (with different CW) after successful transmission, since
it is reset to initial stage 0 after each successful transmission.
GDCF can easily be extended to support priority applications
or QoS differentiation through configuring te
value for dif-
ferent type of applications. A simple method is to let high-pri-
ority applications choose smaller
, while low-priority applica-
tions with larger
. Evidently, the nodes with smaller can seize
1238 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 4, JULY 2004
the channel more quickly and result in lower access delay. This
is particularly important for some real-time multimedia appli-
cations. We will exploit and evaluate this capability of GDCF in
Sections IV and V through simulations.
One issue in GDCF is how to set the parameter
. The intu-
ition is that in the environment with many (or a few) competing
nodes, it requires large (or small) value of
. If the number of
current competing node number can be obtained, we can intelli-
gently adjust the parameter
, but it usually is expensive to pre-
cisely obtain the number of competing nodes in a distributed dy-
namic environment, where nodes consistently move or/and fre-
quently switches on or off. However, we will show that the pa-
rameter
is not too sensitive to the number of competing nodes
in Section IV. In addition, we prove that the short range 4
8
is the optimal value for
if the number of competing nodes is
larger than 10.
IV. GDCF P
ERFORMANCE
ANALYSIS
In this section, we will analyze the performance of GDCF,
discuss how to choose parameter
, and investigate the perfor-
mance of GDCF when supporting priority traffics.
A. Saturation Throughput
First, we will deduce the normalized system throughput
of
GDCF, which is equal to the ratio between “average payload
duration in a slot time” and “average length of slot time,” using
some similar procedures and symbols in [3] and [5]. Let
be
the probability that a transmitted packet collides,
be the prob-
ability that a node transmits in a randomly chosen slot time,
be
the backoff stage,
be the maximal backoff stage [Fig. 2(b)],
be the backoff time slot, be the bidimensional state
of each node,
be the stable probability of state , and
be the one-step transition probability from
state
to state . For convenience, we use CW
and interchangeably in the following discussions.
After every transmission collision, GDCF will back off
(increase the stage
) and double the contention window, so
.
The backoff timer will decrease by 1 if the channel is sensed
idle, so
.
If there are
continuous successful transmissions, GDCF
will decrease the backoff stage
and halve the contention
window; otherwise, the node will stay at the current backoff
stage
and keeps the contention window unchanged. We
can approximate this transition probability as follows.
and
, where and let . Then, we can
easily construct corresponding transition equations of GDCFs
Markov model (see Fig. 4) according to its collision-resolution
process in Fig. 3.
Fig. 3. Collision-resolution process in GDCF.
The nonnull one-step transition probabilities can be com-
puted as
,
,
,
, .
(1)
Let
and we can aggregate the
state
into a single state , so it is easy
to get that
(2)
For each
, also has the relationship shown
in (3) at the bottom of the page.
With (2) and
, (3) can be simplified as
(4)
Because the sum of stationary distribution for all states must
be equal to 1, therefore
(5)
.
(3)
WANG et al.: A NEW COLLISION RESOLUTION MECHANISM TO ENHANCE THE PERFORMANCE OF IEEE 802.11 DCF 1239
Fig. 4. Markov chain model of GDCF.
In (5), can be computed using (2) and is standardized
in 802.11b as follows (for DSSS PHY in 802.11,
):
.
(6)
Replacing (5) with (2) and (6), we can get the value of
in
(7) as
(7)
Then, the probability
that a node transmits in a randomly
chosen slot time can be expressed as
(8)
For convenience of the following discussions, we write the
Markov modeling results of (8) into the following function
:
(9)
In the stationary state, a node transmits a packet with prob-
ability
, so the probability that the transmission is collided
conditioned that there are transmissions can be computed as
follows, because collision must happen if there are at least two
nodes to transmit simultaneously:
(10)
where
is number of competing nodes. Equations (9) and
(10) can be solved by numerical computing methods to obtain
the value of
or . Then, we can get the normalized system
throughput
as
(11)
where
is the probability that there is at least one transmission
in the considered slot time.
is the probability that a transmis-
sion is successful.
and are the average time the channel is
sensed busy because of a successful transmission or collision,
respectively. The
represents the average packet length and
is the duration of an empty slot time. and can be com-
puted as
(12)
and can be computed for basic access mode and RTS/CTS
access mode, respectively, as shown in (13) and (14) at the
bottom of the next page where
is
the packet header,
is the propagation delay, and is the
average length of the longest packet payload involved in a col-
lision. In this paper, all the packets have the same fixed size, so
.
We have calculated the normalized system throughput of
GDCF and DCF according to (11). The results are presented
in Figs. 5 and 6, respectively for basic access and RTS/CTS
access modes. It can be easily observed that the influences on
throughput resulted from factors such as access mode, packet
length, and
value in GDCF. First, for both DCF and GDCF,
RTS/CTS access mode and/or large packet size will bring
1240 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 4, JULY 2004
(a) (b)
Fig. 5. Normalized throughput of basic access mode. (a) Packet length 500 B and (b) packet length 1500 B.
(a) (b)
Fig. 6. Normalized throughput of RTS/CTS access mode. (a) Packet length 500 B and (b) packet length 1500 B.
higher throughput. GDCF can obtain improved performance
for both access modes, but the improved performance under the
basic access mode is much larger. Under the basic access mode,
the improved performance will increase with the increasing
of the packet length, because the effect resulted from lowered
collision probability will be more apparent under long packet
length. On the contrary, under RTS/CTS access mode, when
the packet length is small, the collided slot is more comparable,
so the improved performance will decrease with the increasing
of packet length. As the results in Fig. 5, GDCF with
obtains higher throughput than under the basic access
mode. However, both
4 and 8 obtained nearly the same
throughput under the RTS/CTS mode, so the problem is how
to choose the optimal
for different competing node numbers.
This problem will be further analyzed in Section IV-B.
B. Optimal Value for
It can be seen that the value will heavily influence the
throughput performance. The problem is of which value of
is the most optimal for throughput conditioned that the system
parameters, such as in Table I, are given. We can use the
following method to determine the optimal value of
. Let us
write (11) to the following form:
(15)
In order to maximize
, must be minimal, so let
and we can get the optimal value of as
When is too large, we can use the approximation
(13)
(14)
WANG et al.: A NEW COLLISION RESOLUTION MECHANISM TO ENHANCE THE PERFORMANCE OF IEEE 802.11 DCF 1241
TABLE I
S
YSTEM
PARAMETERS (802.11 DSSS)
Let [ can be computed according to (13) and
(14) for basic access and RTS/CTS access mode, respectively]
be the normalized average collision length in the number of slot
times; then, we can finally obtain the optimal value of
as
(16)
After obtaining the optimal value of
, we can easily obtain
the optimal value of
according to (9) and (10). Because the
optimal value of
is dependent on node number and the nor-
malized average collision length
[see (16)], so the optimal
value of
is also dependent on and . However, we can see
from (13) that
is related to packet length under basic access
mode, so the optimal value of
under the basic access mode
is also dependent on packet length. When the packet length is
large, the collision will cause heavy influence on throughput,
so it is required to choose large value of
. Under the RTS/CTS
access mode, because the normalized average collision length
is independent of packet length, the optimal value of is
also irrelevant to it, so we only present the results of RTS/CTS
access mode in Fig. 7, which also presents the minimal and
maximal value of
to make the GDCF throughput higher than
DCF, where the case of
means that DCF obtains higher
throughput at this time (for example, when node number is equal
to 2 and 4). It can be observed that: 1) When
, GDCF
can improve throughput performance (see the dashed curve la-
beled with “minimal
”), even through the original purpose of
GDCF is to improve performance for large node number and
2) when
, the optimal value of for any node number
is smaller than the minimum of maximal value of ,so
when
, GDCF with any value of in the range of 1
8 will obtain higher throughput than DCF. Fig. 6 conclusively
demonstrates that the optimal
can be obtained in the narrow
range 4
8 and nearly independent of node number when the
node number is larger than 10. This implies that even though
one of the original arguments in GDCF is to improve perfor-
mance for large node number, to our pleasant surprise, this es-
sentially shows that GDCF obtains better performance even if
the node number is small when
is chosen properly. We will
further verify this result through simulations in Section V.
C. Performance Under Priority Traffics
Assume that total competing nodes
can be divided into
priorities or groups. Each group has competing nodes
and is configured with parameter
. It is obvious that the group
with small value of
will has the large probability and that
the node belonging to this group will transmit in a slot time
and have large probability
that the transmissions from this
group is collided. Let the normalized throughput obtained by
group
be . Thus, we can describe each group using the
five-tuple
. In the following, we will calculate
the throughput ratio between any two groups
and .
We can write the results from Markov-modeling into the fol-
lowing function
:
(17)
When the transmission is issued by only a node of group
and all other nodes in this group and other group keep idle, this
transmission will succeed. Otherwise, the transmission will be
collided. So the probability
can be computed as
(18)
For given parameters
and , the numerical results of
and can be obtained through solving (17) and (18). Then,
the throughput ratio between any two groups
and can be
calculated as
(19)
Without loss of generality, it is assumed that
. According to the principle of GDCF, the node with small
has the large probability that it will transmit in a slot time, so
. From (18) we can easily get the following
conclusion:
(20)
Through selecting different combination of
for competing
nodes, GDCF can make the nodes with smaller
obtain lower
MAC access delay (because of lower probability
) and larger
throughput (and corresponding total queuing delay if the traffic
arrival rate is limited). This interesting property in GDCF can
be directly utilized to support differentiated QoS in the MAC
layer. For example, if some node needs supporting real-time ap-
plications , has better channel quality, or has lower energy, we
can let it choose a small value of
to obtain a lower delay of
real-time application, optimal throughput, or higher energy ef-
ficiency and longer system alive time. Although there are some
other QoS-supporting approaches, such as fair queuing in the
MAC layer, GDCF is very simple and flexible. We will further
exploit this property through simulations in Section V.
Although
is assumed to be an integral up to now, can be
also configured to be a real number. Then, GDCF needs only a
minor change, as presented in Fig. 8. However, this change will
not influence the simplicity of GDCF.
In Fig. 8, conSuccNum is the continuous successful transmis-
sions and deficit is the additional parameters used to count the
accumulated deficit quantum in each node. Through using pa-
rameter deficit, GDCF can guarantee that node will halve CW
1242 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 4, JULY 2004
Fig. 7. Optimal
value under the RTS/CTS access mode (when
, which means that DCF is the best choice at this time).
Fig. 8. GDCF algorithm when
is a real number.
once every continuous successful time, on average, even if
is a real number.
V. S
IMULATION RESULTS
This section presents the extensive simulation results of
GDCF. Since the collision probability has more influence
on the basic access mechanism than on the RTS/CTS-based
access mechanism, GDCF is sure to obtain better performance
improvement under the nasic access mechanism than under
the RTS/CTS-based access mechanism as that shown in
Section IV. Therefore, here we only present results using the
RTS/CTS-base access mechanism to observe the performance
improvement in GDCF. The same parameters in Table I are also
used in simulations. The nodes are uniformly distributed in the
100 m
100 m two-dimensional (2-D) square spaces. We use
the shadowing propagation model in order to match the real
environment and assume that the 90% packet can be correctly
received within the distance of 150 m. The main performance
metrics of interest are system throughput, fairness index, RTS
failure ratio, and QoS-supporting capability. The throughput
is used to quantify the throughput gain obtained by GDCF.
We adopt the use of fairness index defined in [12], as it is a
commonly accepted metric. The RTS failure ratio (RFR) can
be used to evaluate the energy cost to transmit packets. If RFR
is large, then the energy cost will be high because more RTS
messages are collided and more energy will be wasted. The
simulation time is selected to be 100 s and all the following
results are the average values obtained in ten simulations. In all
the figures of this section, we use
to represent the 802.11
DCF algorithm, for convenience of comparison with the GDCF
algorithm.
A. Saturation Traffic
In this section, the traffic is configured to saturate the system,
so there are always some competing nodes attempting to
transmit packets. We will investigate the throughput, fairness,
and RTS failure ration in GDCF and DCF, respectively. The
optimal value of
is obtained and compared to the analyzed
results in Section IV.
Experiment A.1—TCP Traffic: In this case, all traffic sources
are TCP (NewReno) flows, whose packet size and window size
are 1460 B and 40 packets, respectively. We collect the RTS
failure ratio, saturation throughput, and fairness for a different
number of competing node.
As analyzed in Section IV, GDCF obtains higher throughput
than DCF and keeps good fairness simultaneously. The results
are presented in Table II and Fig. 9, respectively, for a smaller
node number and large node number. Observing from Fig. 9 and
Table II, GDCF obtains higher throughput and better fairness
than DCF. When is between 4 8, GDCF improves throughput
by about 15%–20% for large node number [Fig. 9(a)] and a bit
for small node number (Table II). GDCF also maintains good
fairness property while obtaining higher throughput. Under
large node number, GDCF has much better fairness than DCF,
especially when
[Fig. 9(b)]. The larger value of (for
example,
), the better fairness will be obtained because
large value of
will make all competing nodes stay in the
same backoff stage with high probability. Although the fairness
of GDCF can be worse than DCF when the node number is
small (Table II), GDCF still has good fairness (
0.9) and this
deterioration is only slight. If the fairness index is smaller than
1, the bandwidth obtained in some nodes will be smaller than
the average bandwidth by
. According to
the definition of fairness index in [11], if the fairness index is
the same, the decreased bandwidth
will be decreased when
the total node number is small. So the fairness deterioration
in GDCF under the small node number will cause slighter
influence on bandwidth sharing than the fairness deterioration
in GDCF under a large node number. Moreover, the total
system throughput and the average bandwidth of each node
under small node is large than that under large node, which
will make the influence resulted from fairness deterioration in
GDCF under small node number more slight.
Fig. 10 presents the results of RTS failure ratio, from which
it can be seen that: 1) GDCF has smaller RTS failure ratio than
DCF for any
value; 2) the RTS failure ratio of GDCF will de-
crease when
value increases, but it will keep at a certain level
(for example, the two curves for
and are nearly
WANG et al.: A NEW COLLISION RESOLUTION MECHANISM TO ENHANCE THE PERFORMANCE OF IEEE 802.11 DCF 1243
TABLE II
T
HROUGHPUT AND
FAIRNESS:S
MALL NODE
NUMBER (
IS FOR
DCF)
Fig. 9. Throughput and fairness-large node number (
is for DCF).
Fig. 10. RTS failure ratio-saturation traffic.
overlapped); and 3) the gain of the RTS failure ratio obtained by
GDCF is more apparent when the node number is large; 4) the
RTS failure ratio will increase when the node number increases
and GDCF decreases the backoff stage by only 1 if and only if
there are
continuous transmissions, so it will keep larger con-
tention window with high probability than DCF and has larger
RTS success ratio independent of node number, as in Fig. 10.
Also, it is reasonable that RTS failure ratio will increase when
decreases or node number increases. The results do show than
GDCF has lower RTS failure ratio than DCF. This means that
nodes in GDCF issue fewer RTS message than DCF for trans-
mitting the same information volume.
Fig. 11. Packet dropping in the MAC layer.
In simulations, we also collect the packet drops in the MAC
layer throughout the whole simulation time and Fig. 11 presents
the results for different number of competing node. It can be
observed that there are few packet drops in the MAC level under
GDCF. On the contrary, DCF cause many packet drops in MAC
level, especially when there are many competing nodes.
Experiment A.2—UDP Traffic: We also change the input
traffic from TCP into UDP and vary the total traffic density
from 0.1 to 1.0. GDCF also obtains the improved throughput.
Because the fairness under each case is close to 0.99 and has
little difference under GDCF and DCF, we only present the
results of system throughput (see Fig. 12). When UDP traffic
1244 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 4, JULY 2004
Fig. 12. System throughput—UDP traffic.
density is smaller than the system saturation throughput, the
system throughput of DCF and GDCF will increase with the
UDP density. After the UPD density overruns the saturation
throughput, the system of DCF and GDCF will nearly keep
constant. Similar to the results in Fig. 10, GDCF also obtains a
bit higher throughput than DCF under this case. It can also be
observed that the throughput under UDP is somewhat smaller
than that under TCP, especially when the node number is equal
to 50. The reason is may be that the factually competing node
number may smaller than the total node number because of
TCP ACK-based flow and congestion-control mechanism.
In summary, GDCF obtains higher throughput and better fair-
ness at the same time. On the selection of the
value, it can
be seen from the above discussions (Table II and Fig. 9) that
GDCF with 1
4 has better performance when the node number
is very small (between 2 and 6) and 4
8 is more suitable
for cases with a large node number. Considering the tradeoff
between throughput decrease (under very small node number)
and throughput and fairness improvement (under large node
number), 4
8 is the better choice if the number of competing
node cannot be known. Recall that the optimal
value (4 8)
from theoretical in Section IV; we can conclude that the simu-
lation results are very consistent with the theoretical results in
Fig. 7, so we can choose
value from 4 8 in practical deploy-
ment.
B. QOS Supporting
Recently, it was shown in 802.11e and enhanced DCF
(EDCF) [8] that it can provide QoS differentiation by con-
figuring small (or large)
and and DIFS for
high- (or low-) priority traffic. In a DCF environment, however,
it is hard to deploy admission control for the high-priority
traffic, so EDCF will cause performance deterioration if most
of applications are high-priority traffics with small
and
value. One of the properties in GDCF is that the node
with smaller
will get the access chances more quickly. We can
use this property directly to support QoS differentiation. In this
section, we investigate the priority-supporting ability in GDCF.
Fig. 13. Experiment B.1—throughput ratio.
Experiment B.1: In this experiment, we divided the total
nodes into two groups with the same node number. The
value in the first group is fixed at , and the for the
second group is varied from 1.1 to 10. UDP traffic is used in
simulations. According to the analysis in Section V-A, the first
group will obtain higher throughput than the second group,
which has larger value of
. We calculate the throughput ratio
between the two groups using (19) and compare it with the
simulation results (see Fig. 13). When
, there is a large
throughput decrease in the second group. When
, the
throughput ratio curve is gradually flat and the increase of
has no apparent effect. Although there is some mismatch
between analysis and simulation results, the first group with
smaller
does obtain higher throughput. On the contrary, the
second group obtains lower throughput. The throughput ratio
between the two groups will decrease with the increasing of
,
so we can adjust the throughput ratio between different types
of traffic in deployment through configuring
.
As for node number
, we also collected instanta-
neous throughput and delay for each group. Fig. 14(a) presents
the instantaneous throughput of group 1 (G1) and group 2 (G2)
and Fig. 14(b) illustrates the instantaneous delay of randomly
chosen nodes from these two groups, respectively. It can be seen
that group 1 with a smaller value of
obtains higher throughput
and lower delay, so GDCF provides a simple and flexible ap-
proach to supporting differentiated QoS.
Experiment B.2: In this experiment, two types of traffic are
configured: 1) high-priority traffic
-UDP source
and 2) low-priority traffic
-TCP source. The
parameters used in TCP are the same as in Section V-A. The
packet size of UDP is equal to 1500 B. The traffic density of
the UDP source is varied from 0.05 to 0.6. In simulations, the
node number is set to 20 and 40, respectively, and one half of
nodes support UDP and the other half support TCP. We collect
the UDP delay, UDP delay jitter, TCP delay, and total system
throughput. Fig. 15 shows that: 1) the average delay and delay
jitter of UDP source in DCF quickly increase with its traffic
density; 2) the average delay and delay jitter of UDP source
in GDCF is not sensitive to the competing node number and
traffic density of UDP source; and 3) UDP source has very
lower delay (0.02–0.035 s) and delay jitter under GDCF than
that under DCF. At the same time GDCF achieves higher system
WANG et al.: A NEW COLLISION RESOLUTION MECHANISM TO ENHANCE THE PERFORMANCE OF IEEE 802.11 DCF 1245
Fig. 14. Experiment B.1—instantaneous throughput and delay.
Fig. 15. Experiment B.2—UDP delay and delay jitter.
Fig. 16. Experiment B.2—Total throughput and TCP delay.
throughput (Fig. 16), but slightly bigger TCP delay than DCF
(Fig. 16).
We have evaluated GDCF through such thee types of
simulations: TCP traffic, UDP traffic, and combined traffic.
The conclusions that GDCF improve better performance than
DCF can be drawn from the extensive simulations. GDCF first
acquires about 15%–20% higher saturation throughput than
DCF. Second, GDCF simultaneously maintains good fairness.
When the number of competing node is some large, GDCF has
better fairness than DCF. Third, GDCF has lower RTS failure
1246 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 4, JULY 2004
ratio, which means that GDCF will issue smaller RTS message
and consumes fewer energy than DCF in order to transmit the
same number of packets. GDCF also drops fewer packets at the
MAC level, while DCF drops many packets in the MAC level,
so GDCF realize better integration of the high-protocol layer
and the low-MAC layer. Finally, GDCF can effectively support
priority traffic. Through simple parameter configuration, GDCF
can make high-priority traffics get much lower delay and delay
jitter and provide differentiated QoS, and keep total higher
throughput at the same time. GDCF needs neither changes
of RTS/CTS structure nor measures of node number; thus, it
is very easy to deploy. The GDCF proposed in this paper is
deterministic approach. It is certain that the behavior of GDCF
can be realized through some probability-based approaches.
For example, we can decrease backoff stage by 1 and halve
the contention window with probability
after each successful
transmission. We also investigate the performance of this
approach and find out that it realizes similar performance with
GDCF and that the optimal value of
is about 0.2. But the fair-
ness of this approach is a bit worse than GDCF because of its
probabilistic behavior in contention resolutions. In future work,
we will plan to evaluate the priority–guarantee mechanism in
GDCF and its performance under multihop environments.
VI. C
ONCLUSION
This paper investigates the MAC protocol for WLAN and
the corresponding collision-resolution algorithm and proposes
an effective algorithm, GDCF, based on 802.11 DCF proto-
cols. Theoretical analysis and simulations are carried out, which
show that the proposed GDCF brings several benefits: 1) it ob-
tains higher throughput than traditional DCF, especially with a
large number of competing nodes; 2) it maintains a good fair-
ness property; 3) GDCF has a lower RTS failure ratio and issues
less RTS messages than DCF in order to transmit the same in-
formation volume, so it is more energy efficient; 4) GDCF drops
fewer packets in the MAC level and can easily extend to support
priority application with the flexibility of selecting different
values; and 5) GDCF is very easy to be deployed, as it does not
need to estimate a competing node number or change the con-
trol message structure and access procedures in DCF.
In our future work, we will investigate the performance of
GDCF when the number of nodes vary frequently and further
study comparison between the performance of GDCF when sup-
porting QoS and 802.11e [8].
R
EFERENCES
[1] A. Chandra, V. Gummalla, and J. O. Limb, “Wireless medium access
control protocols,” IEEE Commun. Surveys Tutorials, vol. 3, no. 2, pp.
2–15, Apr. 2000.
[2] IEEE Standard for Wireless LAN Medium Access Control (MAC) and
Physical Layer (PHY) Specifications, ISO/IEC 8802-11:1999(E), Aug.
1999.
[3] G. Bianchi, “Analysis of the IEEE 802.11 distributed coordination func-
tion,” IEEE J. Select. Areas Commun., vol. 18, pp. 535–547, Mar. 2000.
[4] F. Cali, M. Conti, and E. Gregori, “Dynamic tuning of the IEEE 802.11
protocol to achieve a theoritical throughput limit,” IEEE/ACM Trans.
Networking, vol. 8, pp. 785–799, Dec. 2000.
[5] H. Wu, Y. Peng, K. Long, S. Cheng, and J. Ma, “Performance of reli-
able transport protocol over IEEE 802.11 wireless LAN: Analysis and
enhancement,” in Proc. IEEE INFOCOM’02, vol. 2, June 2002, pp.
599–607.
[6] P. Yong, H. Wu, S. Cheng, and K. Long, “A new self-adapt DCF algo-
rithm,” in Proc. IEEE GLOBECOM’02, vol. 1, Nov. 2002, pp. 87–91.
[7] Y. Kwon, Y. Fang, and H. Latchman, “A novel MAC protocol with fast
collision resolution for wireless LANs,” in IEEE INFOCOM’03, Apr.
2003.
[8] S. Mangold, S. Choi, P. May, O. Klein, G. Hiertz, and L. Stibor,
“IEEE
802.11e wireless LAN for quality of service,” in Proc. Eur. Wireless
(EW’02), vol. 1, Feb. 2002, pp. 32–39.
[9] Y. Xiao, “A simple and effective priority scheme for IEEE 802.11,”
IEEE Commun. Lett., vol. 7, pp. 70–72, Feb. 2003.
[10] Y. E. Sagduyu and A. Ephremides, “Energy-efficient collision resolution
in wireless Ad-Hoc networks,” presented at the IEEE INFOCOM’03,
Apr. 2003.
[11] G. Bianchi and I. Tinnirello, “Kalman filter estimation of the number of
competing terminals in an IEEE 802.11 network,” presented at the IEEE
INFOCOM’03, Apr. 2003.
[12] R. Jain, The Art of Computer Systems Performance Analysis.New
York: Wiley, 1991.
Chonggang Wang received the B.Sc. (Hons.) degree
from Northwestern Polytechnic University, Xi’an,
China, in 1996 and the M.S. and Ph.D. degrees in
communication and information systems from the
University of Electrical Science and Technology,
Chengdu, China and Beijing University of Posts and
Telecommunications, Beijing, China, in 1999 and
2002, respectively.
From September 2002 to Novemeber 2003, he
was an Associate Researcher with the Department
of Computer Science, The Hong Kong University of
Science and Technology, Hong Kong, P. R. China. He currently is a Visiting
Professor with the Special Research Centre for Optical Internet & Wireless
Information Networks (ICOIWIN), ChongQing University of Posts and
Telecommunications (CQUPT), Chongqing, P. R. China, and a Postdoctoral
Research Fellow with the University of Arkansas, Fayetteville.
Bo Li (S’93–M’95–SM’99) received the B.S.
(summa cum laude) and M.S. degrees in computer
science from Tsinghua University, Beijing, P. R.
China, in 1987 and 1989, respectively, and the Ph.D.
degree in electrical and computer engineering from
the University of Massachusetts, Amherst, in 1993.
From 1994 to 1996, he worked on high-perfor-
mance routers and asynchronous transfer mode
(ATM) switches with the IBM Networking System
Division, Research Triangle Park, NC. Since
January 1996, he has been with Computer Science
Department, the Hong Kong University of Science and Technology, where
he is an Associated Professor and Codirector for the ATM/Internet protocol
(IP) cooperate research center, a government-sponsored research center. Since
1999, he has also been an Adjunct Researcher with Microsoft Research Asia
(MSRA), Beijing, China. He has been an Editor or Guest Editor for 16 journals
and is involved in the organization of about 40 conferences. His current research
interests include wireless mobile networking supporting multimedia, video
multicast, and all optical networks using WDM, in which he has published
over 150 technical papers in referred journals and conference proceedings.
Dr. Li was the Co-TPC Chair for IEEE INFOCOM’04 and is a Member of
the Association for Computing Machinery (ACM).
Lemin Li graduated from Jiaotong University,
Shanghai, China, in 1952, majoring in electrical
engineering.
From 1952 to 1956, he was with the Department
of Electrical Communications, Jiaotong University,
Shanghai, P. R. China. Since 1956, he has been with
Chengdu Institute of Radio Engineering (currently
the University of Electronic Science and Technology
of China), Chengdu, P. R. China. From August
1980 to August 1982, he was a Visiting Scholar
with the Department of Electrical Engineering and
Computer Science, University of California, San Diego, where he did research
on digital and spread-spectrum communications. He currently is a Professor of
Communication and Information Engineering. His research work is in the area
of communication networks, including broad-band and wireless networks.
Mr. Li is a Member of the Chinese Academy of Engineering.
