Efficient QoS Differentiation in Crowded Wireless
LANs
Abhik Banerjee, Juki Wirawan Tantra, Chai Kiat Yeo and Bu-Sung Lee
Centre for Multimedia and Network Technology
School of Computer Engineering
Nanyang Technological University
Singapore
Email: abhi0018@ntu.edu.sg, jw.tantra@ieee.org, {asckyeo, ebslee}@ntu.edu.sg
Abstract— The IEEE 802.11e standard has introduced specifications for service differentiation among different classes of data
by specifying four service classes and a new contention resolution
mechanism called EDCA. However, while the protocol shows a
better performance for higher priority data such as voice and
video, the performance is seen to drop drastically at high loads. In
this paper, we explore the effectiveness of a multi-stage contention
scheme for providing QoS differentiation among four different
service classes, as specified by EDCA. From our analysis, we
observe that the multi-stage with prioritization that we propose
gives a much better performance than EDCA for higher priority
data. Moreover, its good performance even at high network loads
shows that this design is much more scalable.1
I. INTRODUCTION
Wireless Local Area Networks (WLANs), with the dominant IEEE 802.11 standard, have been adopted extensively
by mobile users. Recent technology advances have promised
even higher bit rates (> 100 Mbps) in the draft IEEE 802.11n
standard. However, past research have shown that Distributed
Coordination Function (DCF), the MAC layer protocol employed by the IEEE 802.11 standard, is inefficient with a
large number of stations [1]. Xiao et al.[9] have also shown
how the medium access overheads account for throughput and
delay limits 802.11 networks. This inefficiency arises due to
the nonoptimal exponential backoff and contention window
parameters selection. Many methods have been proposed to
increase the efficiency of DCF (e.g. [3] - [6]). Many of
these methods seek to increase the capacity (i.e. throughput
and supported number of users) of DCF by improving the
exponential backoff mechanism. Other mechanisms include
the virtual grouping mechanism suggested in [5] where the
channel is time shared among different groups. In [6], Yang
et al. propose a scheme which involves dividing the stations
into active and non-active groups.
In [2] we introduce a multistage contention (MS) scheme
which is shown to be efficient in a wide range of network
load (from small to large number of stations) without dynamic adaptation mechanism. Briefly described, Multi-stage
contention protocol is as follows. Each contention round is
1 This
work is supported in part by
(a) University Research Committee (Grant No: RG46/06), and
(b) Agency for Science and Technology Research (A*STAR) of Singapore
(Grant No: 0721010028 M47020034 ).
978-1-4244-1645-5/08/$25.00 ©2008 IEEE
1806
divided into multiple stages with only the winners from the
final stage going on to actually transmit data frames. For each
of the previous stages, all the stations wait for duration of
time while their backoff timers expire following which they
transmit a jamming signal. A jamming signal being detected
on the channel signals the end of one stage of contention and
start of the next, in which the winners of the previous contend.
IEEE has introduced a QoS enhancement to DCF, called
Enhanced Distributed Channel Access(EDCA), in the IEEE
802.11e MAC standard. EDCA specifies a set of parameters
to achieve prioritization of data. These include varying the
contention window parameters (CWmax , CWmin ), interframe
spacing (AIFS), and transmission opportunity (TxOP). EDCA
also specifies four different access categories(AC) which
are Voice(AC-VO), Video(AC-VI), Best Effort(AC-BE) and
Background(AC-BK) in order of decreasing priority. The
parameters are configured for each category depending on their
priority.
It is obvious that parameters selection is even more important in EDCA than in DCF and the efficiency of the
protocol plays prominent role in the service quality received
by the users. A number of papers have focussed on developing models to help analyse the efficency of the EDCA
mechanism and how it can be optimized further. Using a
model to analyze the contention window size differentiation
in EDCA, Xiao[9] proposed certain optimizations which can
help improve the performance. Tantra et al. [7] developed
a model which concentrates on the operations of AIFS and
contention window differentiation to analyze the throughput
and delay performance. They also proposed a model to analyze
behaviour of EDCA under statistical traffic [8]. Engelstad et al.
have produced models ([13] - [14] ) which involve analyzing
and predicting the delay in 802.11e.
However, the nature of exponential backoff mechanism,
which characterizes the EDCA, is nonoptimal for crowded
WLANs and slow in adapting to the current network load.
This causes the selected parameters to be suboptimal for either
low or high load. A number of papers seek to provide service
differentiation through other mechanisms apart from EDCA.
In this paper we extend the MS scheme to support traffic
prioritization and show that MS with prioritization (MSP) performs much better than IEEE 802.11e EDCA. There
First Stage
DIFS
Slot
Fig. 1.
Second Stage
Signal
Frame transmission
Illustration of the two-stage contention scheme
are many benefits of the MS-P scheme: (1) MS-P offers a
much higher throughput compared with EDCA; (2) MS-P
provides high throughput for a wide range of load, thus it is
unnecessary to perform dynamic adaptation of the protocol
parameter (e.g., exponential backoff); (3) the simplicity of
MS-P protocol means easy parameters selection; (4) MS-P
eliminates shortterm unfairness issue caused by exponential
backoff; and (5) MS-P is easily extensible for application
specific requirements through parameters adjustment.
The rest of the paper is organised as follows: Section II
describes the Multi-Stage contention Scheme in more detail;
Section III describes the Multi-Stage Contention Scheme with
Prioritization while Section IV analyzes the results and Section
V provides the conclusions.
II. M ULTI -S TAGE C ONTENTION S CHEME
We describe the Multi-Stage(MS) Contention Scheme here
by considering the simplest scheme, which is a two-stage
contention scheme. All the discussions and the analyses in
this paper also consider a two-stage contention scheme which
can be easily extended to more than two stages.
The MS scheme proceeds by dividing the contention into
rounds, where each round is independent of any other round
and all the stations can compete in each round. Each round
is divided into two stages in case of a two-stage contention
scheme. In the first stage, stations choose a backoff counter
from a range (0, W1 − 1) where W1 is the contention window
size for the first stage. The stations listen to the channel while
they wait for their backoff counters to expire, after which they
transmit a short jamming signal2 to end the first stage. As
the jamming signal can be detected by all the stations, the
stations which do not transmit it detect the ones transmitted
by the other stations ans lose the contention round. Thus,
the stations that do transmit the signal win the first stage.
These stations now contend in the second stage to contend
for actually transmitting the data frames. It might be worth
noticing here that unlike MS, in DCF, these stations which
win the first stage would go on to transmit data frames which
might result in collisions.
In the second stage, the stations that win the first stage
again choose a backoff counter in the range (0, W2 − 1). After
waiting for their counters to expire, these stations transmit
their frames. Collisions could still happen in the second stage;
2 The jamming signal is a short period signal that is detectable by the other
stations. Detectable here means that the other stations can detect that there
is a signal in the channel with a fixed constant length. It is unnecessary to
identify or decode the signal.
1807
however, with proper parameters selection, the probability
of successful transmission is much higher compared with
previous protocols such as DCF.
In order to understand this, let us consider a multi-stage
scenario with W1 = W2 = 16 with 20 stations. Thus, each
of the active stations chooses a backoff counter for the first
stage in the range (0, 15). The stations that transmit a jamming
signal after their backoff counters expire win the first stage of
contention. Thus, in our scenario, we may have one or more
winners of the first stage. As mentioned earlier, in case of a
single stage DCF, these stations would transmit data frames
resulting in collisions. These winners then take part in the
second stage of contention and choose backoff counters in
the range (0, 15) since W2 = 16. Once the backoff counter
expires, a station transmits its data frame. As the number of
stations contending were already narrowed down after the first
stage, the possibility of two or more stations transmitting at
once is reduced here, thus resulting in fewer collisions and
higher number of successful transmissions.
The idea of how a station contends in the two-stage contention scheme can be further understood from Fig 1. As
shown here, the station first chooses a backoff counter for
the first stage. Once the station expires its backoff counter, it
transmits a slot length jamming signal and subsequently enters
the second stage. Transmission of the larger data frame takes
place after the station expires the backoff counter it had chosen
for the second stage.
Performance analysis of the two-stage contention protocol
configured with different sets of parameters shows that it
performs better than the standard single stage DCF protocol
with proper parameter selection.
III. M ULTISTAGE C ONTENTION WITH P RIORITIZATION
(MS-P) S CHEME
As with multi-stage contention, MS-P divides the contention
into independent rounds. The stations thus compete in a fair
manner in each round. The EDCA, in contrast, introduces short
term unfairness due to the configuration parameters of different
service classes. As mentioned earlier, each round is divided
into two stages. In the first stage, the stations contend not to
transmit data frames but only to transmit the jamming signal.
Thus, these are the only stations that contend in the second
stage. The stations that win in the first stage now contend in
the second stage to transmit the actual data frames. In MS-P,
the higher priority stations are more likely to transmit their
frames and hence win the contention.
A. Illustration of the protocol
In order to understand how this works, let us consider
W1 and W2 to be the contention windows for the first
and the second stages, respectively. In order to provide QoS
differentiation, we assign a separate set of (W 1, W 2) for
stations of each priority class. In the 802.11e specification,
EDCA specifies four different access categories (AC), namely
AC-VO (voice), AC-VI (video), AC-BK (background), ACBE (best-effort). Accordingly, let {c0; c1; c2; c3} be the set
probability that all slots upto and including the i-th slot are
idle can be computed by
PIi (n) =
i
m (1 − τx,j )nj
(2)
j=1 x=0
where nj is the number of actively contending stations in
category j.
Fig. 2.
Illustration of the two-stage contention scheme with prioritization
of priority classes of the stations with c0 (AC-VO) being the
highest priority class and c3 as the lowest priority (AC-BE). It
is easy to see that W1 is the stronger differentiating factor than
W2; thus, to provide large differentiation among the classes,
W 1(c0) > W 1(c1) > W 1(c2) > W 1(c3) holds. W2 serves
as the second differentiating factor, thus essentially the set
(W 1, W 2) can provide differentiation for a large set of classes.
Fig.2 illustrates a possible scenario with stations of different
categories contending for the channel using the MS-P scheme.
The window sizes for the two stages are as indicated in
the figure for each category. In the first stage, the access
categories AC-VO and AC-VI expire their backoff counters
and transmit jamming signals. The access categories AC-BE
and AC-BK detect the jamming signals transmitted by ACVO and AC-VI but do not transmit any themselves and lose
out the contention after the first stage. Thus, in the second
stage, only the categories AC-VO and AC-VI contend. As the
station with AC-VO has a smaller window size, it has a higher
probability of expiring it’s backoff counter earlier and thereby
winning the contention round, as shown in this case.
Define Ti,j as the random variable of the number of stations
in category i transmitting on the j-th slot, given that none of
the stations have transmitted upto slot i − 1. The probability
density function for Ti,j can be computed as:
P {Ti,1 = t1 , Ti,2 = t2 , . . . , Ti,m = tm }
m nj tj
τi,j (1 − τi,j )nj −tj
=
t
j
j=1
Let Sj be the random variable of the number of stations
winning a contention round from the category j. The probability density function for all the random variables from all the
m categories, S1 , S2 , . . . , Sm is given as fW,n (s1 , s2 , . . . , sm )
and is computed as
fW,n (s1 , s2 , . . . , sm )
B. Analytical Model for MS-P
To derive the throughput of MS-P, first we derive the density
function of the number of winners from each category for
the first stage. Having the density function for the first stage,
then we compute the density function of the second stage
through conditioning of the density function. The success rate
for a particular class equals the probability that the contention
yields one winner from that class and no winner from the
other classes. The throughput for a class c is given by the
success rate of the class divided by the average length of a
transmission round (adjusted with the payload and the frame
lengths).
Let τi,j be the probablity that a station in the j-th category
transmits on the i-th slot in a round. This probability τi,j is
given by
τi,j =
1
Wj − i
(1)
Let us consider L be the set of access categories. For
our discussion, we assume there are m categories. Now, the
1808
(3)
=
Wmin
m
−2 i=0
j=1
nj s j
τ (1 − τi,j )nj −sj PIi−1 (n)
sj i,j
+ αPIWmin −2 (n)
(4)
where α = 0 if s1 = n1 , s2 = n2 , . . . , sm = nm or 0
otherwise. Also, we assume here that PI−1 = 1.
For the two stage contention scheme with prioritization,
let Sj and Rj denote the random variables of the number
of winners from category j in the first stage and the second
stage respectively. Therefore, Rj represents the overall number
of stations in category j which win the two-stage contention
scheme. Considering W1,j and W2,j to be the contention windows for category j of the first stage and the second stage respectively. The density funstion, fR1 ,R2 ,...,Rm (r1 , r2 , . . . , rm )
is given by
30
120
MS with P W1=W2=8, AC-VO
MS with P W1=W2=16, AC-VI
MS with P W1=W2=32, AC-BK
MS with P, Total Throughput
EDCA, AC-VO
EDCA, AC-VI
EDCA, AC-BK
EDCA, Total Throughput
25
MS-P, AC-Voice
MS-P, AC-Video
EDCA, AC-Voice
EDCA, AC-Video
80
Mean Delay (ms)
Throughput (Mbps)
20
100
15
60
10
40
5
20
0
0
16
24
32
40
48
56
64
72
80
88
96
16
24
Number of stations
Fig. 3.
32
40
48
56
64
72
80
88
96
Number of stations
This figure compares the throughput of MS-P vs EDCA.
Fig. 4.
This figure plots the mean delay performance.
frame spaces and σ as an idle-slot duration, the throughput
for a category j, γ2sj is obtained as
fR1 ,...,Rm (r1 , . . . , rm )
=nj
m sj
fR ,...,Rm (r1 = 0, . . . , rj = 1, . . . , rm = 0) · E[P ]
=
P {R1 = r1 , . . . , Rm = rm |S1 = s1 , . . . , Rm = rm }· γ2sj = 1
Tf + (E[I2s ] + 1) · σ
j=1 sj =1
(8)
IV. R ESULTS
P (S1 = s1 , . . . , Sm = sm )
=
=nj
m sj
fW2 ,S1 ,...,Sm (r1 , . . . , rm )fW1 ,n (s1 , . . . , sm )
j=1 sj =1
(5)
The success rate of the system for a particular category equals the probability that the contention yields one
winner from that category, i.e.fR1 ,...,Rm (r1 = 0, . . . , rj =
1, . . . , rm = 0). We need to compute the mean idle duration
in each contention round in order to obtain the throughput of
the system.
The mean idle duration for the two stages are expressed as
E[I1 ] and E[I2 ]:
W1,min −1
E[I1 ] =
i(1 − (1 −
i=0
E[I2 ] =
nj
m j=1 sj =0
i=0
(1 − τi,j )nj )PIi−1
(6)
j=1
fW1j ,n (s1 , . . . , sm )·
W2,min −1
m
i(1 − (1 −
m
(1 − τi,j )nj )PIi−1
(7)
j=1
The total idle duration for the two-stage contention scheme,
E[I2s ], is given as E[I2s ] = E[I1 ] + E[I2 ].
With E[P ] as the mean payload size, Tf as the mean
transmission time including the acknowledgement and iner-
1809
We have analyzed the throughput and mean delay performances for the MS-P as compared to EDCA. We consider
the same set of access categories as specified in the EDCA
specifications. Hence, L = {0, 1, 2, 3} is the set of 4 access
categories.
Fig. 3 compares the throughput of MS-P with that of EDCA.
The symbols represent simulations results whereas the lines
represent the analytical results computed using the formulas.
If we compare the throughput offered by MS-P to that offered
by EDCA, the former performs much better consistently for
all categories of data.
What is especially noticeable is that at very high loads
with 50 or more users contending, MS-P offers throughput
twice that of EDCA for the access category AC-VI. Also, as
mentioned earlier, the performance doesn’t vary by much as
the load increases, thus indicating a scenario wherein voice
sessions don’t experience a degradation of service even when
the network gets congested.
MS-P also provides a better throughput for the AC-VI at
high loads, though the throughput is similar at lower loads.
However, the total throughput given by the MS-P scheme is
much higher than that of the EDCA. Again, at very high loads,
MS-P has a throughput of about 300% of that of EDCA.
In EDCA, every station determines when to transmit data
using a contention resolution mechanism involving a backoff
counter determined by the maximum and minimum window
sizes. The station then waits for the duration of the AIFS
before finally transmitting a frame. While configuring these
parameters can provide service differentiation, it isn’t highly
effective in reducing the possibility of collisions during data
transmission. The multi-stage scheme helps reduce this possibility by increasing the chances of ending with a single winner.
With only a few stations contending to actually transmit data
in the second stage, there is a higher chance that only one will
actually transmit the data.
Fig. 4 compares the mean delay performances of MS-P to
that of EDCA for the two highest categories of data. The mean
delay for Voice category data (AC-VO) is much smaller for
MS-P. At very high loads, the delay for voice data is less
than half of the mean delay in case of EDCA. This could be
attributed to the smaller window sizes for higher priority data
such as voice. As the window sizes for both stages are smaller,
in the case of stations with higher category data winning
the contention, the stages get over earlier. Thus, the delay
encountered before transmitting a data frame is considerably
lower. Also, as explained earlier, a higher chance of collision
free data transmissions also implies fewer retransmissions and
hence retransmission delays.
The mean delay for AC-VI also shows a reduction by about
40% for MS-P at high loads though it is comparable to EDCA
at light loads.
Thus, MS-P can be seen to be a more scalable scheme
with better support for service differentiation among different
categories of data. MS-P scores over EDCA primarily because
of it’s effectiveness in reducing the number of colisions after
each contention round. Further, the extra overheads in EDCA
such as the AIFS contribute to a lower throughput as compared
to the MS-P.
V. C ONCLUSION
Our proposed MS-P scheme provides QoS differentiation
for stations of different classes through the differentiation of
the contention window set (W1, W2). Our analysis shows that
MS-P provides much higher throughput compared with EDCA
especially with high network load. We have illustrated the
multiple benefits of MS-P. To be highlighted is the capability
of MS-P to provide shortterm fairness, which is difficult to
achieve with the standard exponential backoff procedure. We
conclude that MS-P is a viable technology providing QoS
support that is simple in operations and capable to perform
well with a wide range of network load especially in a crowded
WLANs.
1810
Future additions to the scheme can include further enhancements to the scheme which can help address scenarios where
the presence of hidden nodes in the network cause a hindrance
to the overall network throughput.
R EFERENCES
[1] G. Bianchi, ”Performance analysis of the IEEE 802.11 distributed coordination function,” IEEE J. Select. Areas Commun., vol. 18, no. 3, pp.
535547, Mar. 2000.
[2] Juki Wirawan Tantra, Teck Meng Lim, BuSung Lee, and Chai Kiat Yeo,
”Effectiveness of explicit stations grouping in crowded wireless LANs”
IEEE Commun. Lett., 2007.
[3] C.Wang, B. Li, and L. Li, ”A new collision resolution mechanism
to enhance the performance of IEEE 802.11 DCF,” IEEE Trans. Veh.
Technol. , vol. 53, no. 4, pp. 1235 1246, July 2004.
[4] Y. Kwon, Y. Fang, and H. Latchman, ”Design of MAC protocols with
fast collision resolution for wireless local area networks,” IEEE Trans.
Wireless Commun., vol. 3, no. 3, pp. 793807, May 2004
[5] S.-M. Kim and Y.-J. Cho, ”Performance evaluation of IEEE 802.11
distributed coordination function with virtual group,” IEICE Trans. Commun.,vol. E88-B, no. 12, pp. 46254635, Dec. 2005.
[6] X. Yang and N. H. Vaidya, ”A wireless MAC protocol using implicit
pipelining,” IEEE Trans. Mobile Comput., vol. 5, no. 3, pp. 258273, MayJune 2006.
[7] J.W. Tantra, Chuan Heng Foh, and A.B. Mnaouer, ”Throughput and
Delay Analysis of the IEEE 802.11e EDCA Saturation”, IEEE International Conference on Communications, 16-20 May 2005, Volume 5,
Page(s):3450 - 3454.
[8] J.W. Tantra, Chuan Heng Foh, Tinnirello, I., Bianchi, G., ”Analysis of
the IEEE 802.11e EDCA Under Statistical Traffic”, IEEE International
Conference on Communications, Volume 2, June 2006 Page(s):546 - 551
[9] Y. Xiao, ”Performance analysis of IEEE 802.11e EDCF under saturation
condition,” in Proc. IEEE ICC, Paris, France, June 2004.
[10] Y. Xiao and J. Rosdahl, ”Throughput and delay limits of IEEE 802.11,”
IEEE Commun. Lett., vol. 6, no. 8, pp. 355357, Aug. 2000
[11] Tiantong You, Chi-Hsiang Yeh, Hassanein, H., ”DRCE: a high throughput QoS MAC protocol for wireless ad hoc networks,” Proceedings of the
10th IEEE Symposium on Computers and Communications (ISCC 2005)
[12] Zhang Wei, Sun Jun, Liu Jing and Zhang Hai-bin, ”Performance
analysis of IEEE 802.11e EDCA in wireless LANs,” Journal of Zhejiang
University SCIENCE A, 2007 8(1):18-23
[13] Engelstad, P.E. and ØsterbøO.N., ”Analysis of the Total Delay of IEEE
802.11e EDCA and 802.11 DCF”, Proceedings of IEEE ICC 2006
[14] Engelstad, P.E., ØsterbøO.N., ”Delay and Throughput Analysis of IEEE
802.11e EDCA with AIFS Differentiation under Varying Traffic Loads”,
Proceedings of the Fifth International IEEE Workshop on Wireless Local
Networks (WLN 05), Sydney, Australia, Nov. 15-17, 2005.
[15] IEEE Std. 802.11e, Part 11: Wireless LAN Medium Access Control
(MAC) and Physical Layer (PHY) specifications Amendment 8: Medium
Access Control (MAC) Quality of Service Enhancements, Nov 2005