Performance Anomaly of 802.

11b
Martin Heusse, Franck Rousseau, Gilles Berger-Sabbatel, Andrzej Duda

LSR-IMAG Laboratory
Grenoble, France
email: {heusse, rousseau, gberger, duda}@imag.fr

Abstract We analyze the performance of the IEEE 802.11b the level of the lower rate. Such a behavior penalizes fast hosts
wireless local area networks. We have observed that when some and privileges the slow one. The reason for this anomaly is
mobile hosts use a lower bit rate than the others, the performance the basic CSMA/CA channel access method which guarantees
of all hosts is considerably degraded. Such a situation is a
common case in wireless local area networks in which a host far that the long term channel access probability is equal for all
away from an Access Point is subject to important signal fading hosts. When one host captures the channel for a long time
and interference. To cope with this problem, the host changes its because its bit rate is low, it penalizes other hosts that use the
modulation type, which degrades its bit rate to some lower value. higher rate.
Typically, 802.11b products degrade the bit rate from 11 Mb/s to In this paper, we analyze the performance of the 802.11b
5.5, 2, or 1 Mb/s when repeated unsuccessful frame transmissions
are detected. In such a case, a host transmitting for example at 1 DCF access method by deriving simple expressions for the
Mb/s reduces the throughput of all other hosts transmitting at 11 available throughput (Section II). Then we extend the analysis
Mb/s to a low value below 1 Mb/s. The basic CSMA/CA channel to the case of hosts with different bit rates (Section III). We
access method is at the root of this anomaly: it guarantees an validate the analysis by means of simulation (Section IV) and
equal long term channel access probability to all hosts. When measurements. Section V provides several performance mea-
one host captures the channel for a long time because its bit
rate is low, it penalizes other hosts that use the higher rate. We surements and comparisons to illustrate the anomaly. Finally,
analyze the anomaly theoretically by deriving simple expressions we briefly discuss related works (Section VI) and present some
for the useful throughput, validate them by means of simulation, conclusions (Section VII).
and compare with several performance measurements.
II. P ERFORMANCE OF IEEE 802.11 B DCF ACCESS
I. I NTRODUCTION METHOD
We focus our paper on wireless local area networks such The IEEE 802.11b standard [1] defines two access meth-
as IEEE 802.11 that have become popular as access networks ods: the Distributed Coordination Function (DCF) that uses
to the wireless mobile Internet. They can be deployed in hot CSMA/CA to allow for contended access to the wireless
spots areas and offer performance comparable to wired local media and the Point Coordination Function (PCF) providing
area networks. The question of the performance effectively uncontended access via arbitration by a Point Coordinator,
perceived by mobile hosts becomes increasingly important which resides in the Access Point. The DCF method provides
as many new emerging applications such as mobile informa- a best effort type of service whereas the PCF guarantees a
tion access, real-time multimedia communications, networked time-bounded service. Both methods may coexist, a contention
games, immersion worlds, cooperative work require sufficient period following a contention-free period. PCF would be
bandwidth. especially well suited for real-time traffic as it permits to allo-
Although IEEE 802.11b provides a means for allocating a cate the radio channel according to applications requirements,
part of the radio channel bandwidth to some hosts (PCF - but the PCF method is not implemented in current 802.11
Point Coordination Function), the commonly available access products.
method (DCF - Distributed Coordination Function) uses the The DCF access method is based on the CSMA/CA prin-
CSMA/CA protocol to share the radio channel in a fair way. ciple in which a host wishing to transmit senses the channel,
However, we have observed that in some common situations waits for a period of time (DIFS Distributed Inter Frame
in a wireless environment, the method results in a considerable Space) and then transmits if the medium is still free. If the
performance degradation. In a typical wireless local area packet is correctly received, the receiving host sends an ACK
network, some hosts may be far away from their access point frame after another fixed period of time (SIFS Short Inter
so that the quality of their radio transmissions is low. In this Frame Space). If this ACK frame is not received by the sending
case current 802.11b products degrade the bit rate from the host, a collision is assumed to have occurred. The sending host
nominal 11 Mb/s rate to 5.5, 2, or 1 Mb/s when a host attempts to send the packet again when the channel is free for
detects repeated unsuccessful frame transmissions, it decreases a DIFS period augmented of a random amount of time.
its bit rate. If there is at least one host with a lower rate, a Let first consider that a single host in a 802.11b cell
802.11 cell presents a performance anomaly: the throughput transmits a single data frame. If we neglect propagation times,
of all hosts transmitting at the higher rate is degraded below the overall transmission time is composed of the transmission

0-7803-7753-2/03/$17.00 (C) 2003 IEEE 836

 ttr

C
MAC header 4 bytes
Data R
30 bytes C

50 s tcont tpr 10 s tpr tack

ACK
MPDU
14 bytes

DIFS backoff PLCP PLCP SIFS PLCP PLCP

preamble header preamble header

Fig. 1. Successful transmission of a single frame

time and a constant overhead: Collisions and the exponential backoff mechanism influence
the performance of 802.11b for an increasing number of hosts.
T = ttr + tov (1)
The overall frame transmission time experienced by a single
where the constant overhead host when competing with N 1 other hosts has to be
increased by some interval tcont that accounts for the time
tov = DIF S + tpr + SIF S + tpr + tack
spent in contention procedures. So the overall transmission
is composed of the PLCP (Physical Layer Convergence Pro- time becomes:
tocol) preamble and header transmission time tpr , SIF S =
10 s, tack is the MAC acknowledgment transmission time T (N ) = ttr + tov + tcont (N ) (3)
(10 s if the selected rate is 11 Mb/s, as the ACK length is The analytical formula for tcont (N ) is difficult to derive
112 bits), and DIF S = 50 s. ttr is the frame transmission (see [2]) and we propose to use a simple approximation.
time (see Figure 1). tpr varies according to the bit rate used by Considering that the hosts always sense a busy channel when
the host. When it transmits at 1 Mb/s, the long PLCP header they attempt to transmit and that the number of transmissions
is used and tpr = 192 s. If it uses 2, 5.5 or 11 Mb/s, then that are subject to multiple successive collisions is negligible
tpr = 96 s (short PLCP header). For bit rates greater than 1 we find:
Mb/s and the frame size of 1500 bytes of data (MPDU of total
1 + Pc (N ) CWmin
1534 bytes), proportion p of the useful throughput measured tcont (N )  SLOT
above the MAC layer will be: 2N 2
ttr 1500 where Pc (N ) is the proportion of collisions experienced for
p= = 0.70. (2) each packet successfully acknowledged at the MAC level (0

T 1534
Pc (N ) < 1).
So, a single host sending long frames over a 11 Mb/s radio
A simple expression for Pc (N ) can be derived by consider-
channel will have a maximum useful throughput of 7.74 Mb/s.
ing that a host attempting to transmit a frame will eventually
If there are multiple hosts attempting to transmit, the chan-
experience a collision if the value of the chosen backoff
nel may be sensed busy and hosts enter a collision avoidance
interval corresponds to the residual backoff interval of at
phase: a host executes the exponential backoff algorithm
least one other host. Such an approximation holds if multiple
it waits for a random interval distributed uniformly between
successive collisions are negligible. So we have
[0, CW ] SLOT . The congestion window CW varies be-
tween CWmin = 31 and CWmax = 1023, the value of SLOT Pc (N ) = 1 (1 1/CWmin )N 1 . (4)
is 20 s (these parameters are for 802.11b). The host that
chooses the smallest interval starts transmitting and the others Finally, proportion p of the useful throughput that can be
hold counting down until the transmission is over. Each time obtained by a host depends on the number of hosts and is
a host happens to collide, it doubles CW up to CWmax . given by:
In the case of a single host that tries to transmit a sustained p(N ) = ttr /T (N ) (5)
traffic (the host has always a packet to send), the carrier
sense applies also to the hosts own transmissions, so that A. Discussion
it inserts a random interval between each transmission. This The analysis in this section shows that the throughput that
is mandatory, because transmitting frames continuously would can be obtained over the 802.11b WLAN is much smaller than
prevent any other host from accessing the channel. Thus, Eq. 2 the nominal bit rate of 11 Mb/s, for example, if there are no
gives an upper bound that can be attained only for a host collisions, one host may expect the maximum throughput of
passing packets to the MAC layer at the moment the previous 7.74 Mb/s. Furthermore, the proportion of the useful through-
transmission is completed. put strongly depends on the number of competing hosts.

837
 III. P ERFORMANCE ANOMALY OF IEEE 802.11 B A. Discussion
Consider now the situation in which N hosts of different The result of this section show that when one host that uses
bit rate compete for the radio channel: N 1 hosts use the a lower bit rate competes with other hosts, the throughput
high transmission rate R = 11 Mb/s and one host transmits of all hosts may be significantly limited the fast hosts see
at a degraded rate r = 5.5, 2, or 1 Mb/s. In this case, the their throughput decreased roughly to the order of magnitude
frame transmission time depends on the rate: ttr = sd /R or of the slow hosts throughput. When fast and slow hosts share
ttr = sd /r, where sd is the data frame length in bits. The the radio channel of a 802.11b cell, the throughput is bounded
MAC layer ACK frame is also sent at the rate that depends by N X instead of R pf (N ) as we may expect in presence
on the host speed, thus we denote by tR r
ov and tov the associated of a large number of fast hosts.
overhead time.
The fair access to the channel provided by CSMA/CA
Let Tf be the overall transmission time of a fast host
causes a slow host transmitting at 1 Mb/s to capture the
transmitting at rate R:
channel eleven times longer than hosts emitting at 11 Mb/s.
sd
T f = tR
ov + + tcont . This degrades the overall performance perceived by the users
R
in the considered cell, and this anomaly holds whatever is the
Similarly, let Ts be the corresponding time for a slow host
proportion of slow hosts.
transmitting at rate r:
sd
Ts = trov + + tcont .
r B. UDP traffic
We can analyze the overall performance by assuming that
To be able to compare the analytical expressions with
the hosts alternate transmissions: channel utilization by a fast
measurements, we should consider an application level sce-
host is the ratio between the time to send one packet (Tf ) and
nario that governs the traffic generated at the MAC level.
the time during which all other hosts transmit once with all
We consider two basic scenarios. In the first one, N hosts
possible collisions. This holds on a long term, because the
(one of these hosts may send at the lower rate) generate
long term channel access probability CSMA/CA is equal for
a connectionless UDP stream to a host behind the Access
all hosts. However, this is not true for a short term it has
Point. As the streams are not acknowledged, the only traffic
been shown that the short term behavior of CSMA/CA is not
generated at the MAC level is composed of UDP data packets.
fair [3]. So, we will have
In this case, the throughput experienced by each host will be
Tf given by Eq. 7.
Uf = (6)
(N 1) Tf + Ts + Pc (N ) tjam N
where tjam is the average time spent in collisions. The duration
C. TCP traffic
of a collision depends on the type of the hosts involved in the
collision (slow or fast). tjam can be found by considering all In the second scenario, there are N 1 hosts that send data
possible pairs between N 1 fast hosts and a slow one (see over TCP connections to a server behind the Access Point.
Appendix): We assume that the server is connected to the Access Point
2 2 via a 100 Mb/s switched Ethernet, so that the wireless link is
tjam = Ts + (1 ) Tf .
N N the only bottleneck. In this configuration, TCP data segments
The throughput at the MAC layer of each of the N 1 fast and ACKs travel only once over the wireless channel. As data
hosts is segments are acknowledged by the server, we have a total of
Xf = Uf pf (N ) R N hosts competing for the radio channel (N 1 hosts and the
where pf (N ) = sd
. Finally, we obtain Access Point that forwards TCP ACKs from the server).
R Tf
sd We denote by Tfack (resp. Tsack ) the time to transmit a frame
Xf = (7) containing a TCP ACK to a fast (resp. slow) host at rate R
(N 1) Tf + Ts + Pc (N ) tjam N
(resp. r). In the TCP scenario, N 2 hosts send their packets
Similarly, we can express the channel utilization of the slow at higher rate R of 11 Mb/s, one host sends at lower rate r
host as (1, 2, 5.5 Mb/s), and the Access Point sends TCP ACKs at a
Ts variable rate that depends on the rate of the host which is the
Us = (8)
(N 1) Tf + Ts + Pc (N ) tjam N destination of the ACK. As TCP sends ACKs for every other
so that the throughput at the MAC layer of the slow host is data segment transmitted (we have effectively observed such
a behavior in our experiments), the throughput in the TCP
Xs = Us ps (N ) r
scenario can be expressed as:
sd
where ps (N ) = r Ts .
Finally, we obtain the following:
Result. The fast hosts transmitting at the higher rate R obtain sd
X= Tfack Tsack
the same throughput as the slow host transmitting at the lower (N 2)(Tf + 2 ) + Ts + 2 + Pc (N ) tjam 2 (N
3
1)
rate r: (10)
Xs = Xf = X. (9) The derivation of tjam is given in Appendix.

838
 IV. S IMULATION
7
To validate the analytical expressions given in the previous 11 Mb/s - 11 Mb/s
Eq. 7: 11 Mb/s - 11 Mb/s
section and to gain insight into the complex backoff mech- 6 1 Mb/s - 11 Mb/s
anism of CSMA/CA we have developed a simple 802.11b Eq. 7: 1 Mb/s - 11 Mb/s

Throughput (Mbits/s)
5.5 Mb/s - 11 Mb/s
simulator. It simulates the CSMA/CA access method under 5 2 Mb/s - 11 Mb/s

heavy load, i.e. we assume that the channel is always sensed

4
busy when hosts need to transmit. In this way we can evaluate
a worst case bound for Pc (N ). All the parameters of the sim- 3
ulation such as CWmin , CWmax , SLOT have values defined
in the 802.11b standard. 2
Figure 2 shows the simulated value of Pc (N ) for a varying
number of hosts and compares it with Eq. 4 as well as with 1

several measured values. Each point of the graph comes from

0
the simulation of 1000000 transmitted packets. 0 2 4 6 8 10 12 14 16 18 20
Number of hosts

Fig. 3. Throughput experienced by a 802.11b host when all hosts except

60 one transmit at 11Mb/s
Simulation
Eq. 4
50 Experimental

Average throughput all 11 Mb/s

40 10 Cumulative throughput all 11 Mb/s
Average throughput all 2 Mb/s
Cumulative throughput all 2 Mb/s
Throughput (Mbits/s)

Average throughput one 2 Mb/s

Pc

30 8 Cumulative throughput one 2 Mb/s

20
6

10
4

0
0 2 4 6 8 10 12 14 16 18 20 2
Number of hosts

Fig. 2. Proportion of collisions Pc for a varying number of hosts 0

0 2 4 6 8 10 12 14 16 18 20
Number of hosts
As shown in the previous section, all hosts obtain the same
throughput even if one of them transmits at a lower rate. The Fig. 4. Cumulative throughput in a cell with a single slow host
throughput is shown in Figure 3 for different rates of the
slow host and for a varying number of hosts in the cell. The
throughput curves obtained from simulation or expression 7 V. P ERFORMANCE MEASUREMENTS
are superimposed. We have set up a platform to measure the throughput that
Finally, we give the cumulative throughput of all hosts. hosts can obtain when sharing a 11 Mb/s 802.11b wireless
Although the presence of a slow host decreases the throughput cell. We have used four notebooks (Marie, Milos, Kea, and
perceived by the other ones, its impact on the cumulative Bali) running Linux RedHat 7.3 (kernel 2.4.18) with 802.11b
throughput decreases when the number of hosts increases (cf. cards based on the same chipset (Lucent Orinoco and Compaq
Figure 4). This is due to the fact that the slow host uses a WL 110). The wired part of the network is connected by a
diminishing proportion of the channel as the number of hosts Lucent Access Point. The notebooks use the Wvlan driver for
rises. the wireless cards. The cards do not use the RTS/CTS option
that may optimize performance in case of the hidden terminal
A. Discussion problem.
Our simulation results show that the precision of the an- We measure the throughput using three tools:
alytical expressions is fairly good. The expression for the netperf generates TCP or UDP traffic to a target host
proportion of collisions slightly overestimates the impact of running netserver and measures the throughput at the
collisions. However, this discrepancy results in a small error in end of the session [4].
the throughput formulae, because the proportion of collisions tcpperf generates TCP traffic and measures the
is only a factor of the average time spent in collisions tjam . throughput obtained during each second.

839
 udpperf generates UDP traffic and measures the

throughput obtained during each second. 12000
Bali
The measurements are done using netperf and compared Marie
Rate Bali
with the results of tcpperf and udpperf. We generate all 10000

Throughput (Kbits/s)
traffic to a host on the wired part of the network.
8000
A. Hosts with different rates, no mobility
In the first experiment, we place all notebooks near the 6000
Access Point to obtain good transmission conditions on the
radio channel and we force one of them to work at a degraded 4000
bit rate. We measure the throughput for a varying number of
hosts and different traffic conditions. 2000
Table I compares the measured throughput with the expres-
sions derived in Section III when hosts generate TCP streams. 0
They compete with the Access Point that sends TCP ACKs on 0 50 100 150 200 250 300

behalf of the destination (N = 2, 3, 4, 5). The table references Time (secs)

figures that present the evolution of the throughput in time for Fig. 6. Measured throughput in Mb/s for two hosts, UDP traffic
a given number of hosts.
Table 10 compares the throughput obtained in the same
setup with UDP streams. The hosts compete with each other 12000
for the radio channel (N = 1, 2, 3, 4). Bali
Marie
10000 Milos
Rate Bali
Throughput (Kbits/s)

12000
Bali 8000
Marie
10000 Rate Bali
6000
Throughput (Kbits/s)

8000
4000

6000
2000

4000
0
0 50 100 150 200 250 300
2000 Time (secs)

Fig. 7. Measured throughput in Mb/s for three hosts, TCP traffic

0
0 50 100 150 200 250 300
Time (secs)
C. Hosts with different rates, real mobility
Fig. 5. Measured throughput in Mb/s for two hosts, TCP traffic
In the second experiment, Bali is a mobile host that moves
around and its bit rate automatically adapts to varying trans-
B. Discussion mission conditions. The other hosts are located near the Access
We can observe that the measured values correspond fairly Point and their transmission conditions are good.
well to the analytical expressions. The precision of the ana- Figure 11 shows the evolution of the throughput in time
lytical expression is better for the UDP traffic than for that when two hosts send TCP traffic. Marie keeps its bit rate at
of TCP. In the case of TCP, the traffic pattern is much more 11 Mb/s and Bali changes its bit rate according to the quality
complex than that of UDP, because the Access Point competes of the radio channel (cf. Figure 12). They compete with the
with other hosts when sending TCP ACKs. In addition to that, Access Point that sends ACKs on behalf of the destination
the traffic of TCP ACKs is governed by a complex dependence (N = 3).
on the overall round trip time and the bottleneck link between It can be observed that when transmission conditions are
the source and the destination. It can become correlated with bad (period 300-380) the throughput of Marie increases. This
the traffic of data segments a TCP ACK is sent upon arrival can be explained by the fact that in adverse conditions the TCP
of a data segment. source of Bali limits its sending rate so that Marie may benefit
The influence of the estimate for the proportion of collisions from the unused channel capacity. This situation becomes
on the overall throughput is limited, because it only accounts more explicit after instant 380 when Bali stops sending even
for a small factor in Eq. 7. if its bit rate is not zero while Marie gains almost all the

840
 Bit rate of Bali Bali Marie Milos Kea Eq. 10 observed Pc Eq. 4 Figure No.
11 5.08 - - - 7.21 4% 3.1% -
5.5 3.37 - - - 4.12 4% 3.1% -
2 1.55 - - - 1.65 4% 3.1% -
1 0.83 - - - 0.83 4% 3.1% -
11 2.52 2.48 - - 3.48 8% 6.2% 5
5.5 2.04 1.96 - - 2.51 8% 6.2% 5
2 1.12 1.11 - - 1.27 8% 6.2% 5
1 0.67 0.63 - - 0.70 8% 6.2% 5
11 1.65 1.54 1.52 - 2.24 8% 9.1% 7
5.5 1.36 1.39 1.47 - 1.77 8% 9.1% 7
2 0.83 0.89 0.95 - 1.02 8% 9.1% 7
1 0.57 0.52 0.64 - 0.62 8% 9.1% 7
11 1.15 0.92 1.21 1.02 1.63 9% 12.0% 9
5.5 1.16 0.83 1.11 0.83 1.35 9% 12.0% 9
2 0.74 0.41 0.65 0.55 0.85 9% 12.0% 9
1 0.56 0.32 0.41 0.27 0.53 9% 12.0% 9

TABLE I
M EASURED THROUGHPUTS IN M B / S FOR A VARYING NUMBER OF HOST, TCP TRAFFIC

Bit rate of Bali Bali Marie Milos Kea Eq. 7 observed Pc Eq. 4 Figure No.
11 3.09 3.36 - - 3.26 4% 3.1% 6
5.5 2.38 2.42 - - 2.44 4% 3.1% 6
2 1.30 1.26 - - 1.29 4% 3.1% 6
1 0.76 0.76 - - 0.73 4% 3.1% 6
11 2.26 2.0 2.23 - 2.20 6% 6.2% 8
5.5 2.01 1.56 1.89 - 1.77 6% 6.2% 8
2 1.17 0.90 1.16 - 1.06 6% 6.2% 8
1 0.74 0.58 0.69 - 0.63 6% 6.2% 8
11 1.71 1.41 1.8 1.41 1.64 10% 9.1% 10
5.5 1.66 1.16 1.59 1.19 1.38 10% 9.1% 10
2 0.96 0.84 0.81 0.72 0.89 10% 9.1% 10
1 0.69 0.47 0.63 0.49 0.56 10% 9.1% 10

TABLE II
M EASURED THROUGHPUTS IN M B / S FOR A VARYING NUMBER OF HOSTS , UDP TRAFFIC

available throughput. that many papers use simulation [5], [6]. Only a few papers
Figure 13 shows a similar graph when the two hosts send analyze 802.11 analytically using Markov chains [2], [7]. This
UDP traffic (N = 2). Marie keeps its bit rate at 11 Mb/s and approach models the complex behavior of 802.11 better, but
Bali changes its bit rate according to the quality of the radio usually gives complex results or requires numerical resolution.
channel (cf. Figure 14). Short-term unfairness of CSMA-based medium access pro-
The results are similar to the TCP traffic, but we can observe tocols has also been a topic of interest. In [3] the authors
that now Marie does not gain as much throughput when the use experimental and analytical methods to evaluate the short-
transmission conditions are bad (period 300-380) unless Bali term fairness degree of the CSMA-CA as implemented in the
stops sending data at instant 420. WaveLAN network.
VI. R ELATED WORK VII. C ONCLUSIONS
Many papers have studied the performance of 802.11 We have analyzed the performance of the IEEE 802.11b
WLANs, however all of them assume that all hosts commu- wireless local area networks. This analysis shows that the
nicate using the same bit rate. The analysis is difficult so throughput of the 802.11b WLAN is much smaller than the

841
 12000 12000
Bali Bali
Marie Marie
10000 Milos 10000 Milos
Rate Bali Kea
Throughput (Kbits/s)

Throughput (Kbits/s)
Rate Bali
8000 8000

6000 6000

4000 4000

2000 2000

0 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300 350
Time (secs) Time (secs)

Fig. 8. Measured throughput in Mb/s for three hosts, UDP traffic Fig. 10. Measured throughput in Mb/s for four hosts, UDP traffic

12000 6000
Bali Bali
Marie Marie
10000 Milos 5000
Kea
Throughput (Kbits/s)
Throughput (Kbits/s)

Rate Bali
8000 4000

6000 3000

4000 2000

2000 1000

0 0
0 50 100 150 200 250 300 0 60 120 180 240 300 360 420 480
Time (secs) Time (secs)

Fig. 9. Measured throughput in Mb/s for four hosts, TCP traffic Fig. 11. Measured throughput in Mb/s for two hosts (one mobile), TCP
traffic

nominal bit rate. Furthermore, the proportion of the useful deployment of 802.11 access points. The question becomes
throughput strongly depends on the number of competing important in the case of hot spots that cover areas with an
hosts. important number of hosts. When access points are located so
When mobile hosts move, they may encounter bad trans- that some mobile hosts are far away and use a smaller bit rate,
mission conditions and degrade the bit rate from 11 Mb/s this will result in the performance degradation perceived by
to 5.5, 2, or 1 Mb/s. We have analyzed how a host with a all hosts.
lower bit rate influences the throughput of other hosts that
share the same radio channel. Our analysis and performance VIII. ACKNOWLEDGMENTS
measurements show that the slow host may considerably limit This work has been supported by the French Ministry of
the throughput of other hosts roughly to the level of the lower Industry, National Network of Telecommunication Research
rate. (RNRT) via the @IRS++ project.
However, this adverse performance effect should be alle-
viated by the observation that in real conditions and for TCP R EFERENCES
traffic, the host that degrades its bit rate will be anyway subject [1] ANSI/IEEE, 802.11: Wireless LAN Medium Access Control (MAC) and
to important packet losses which in turn limit its sending rate. Physical Layer (PHY) Specifications, 2000.
In this way, other hosts may benefit from the unused capacity. [2] F. Cali, M. Conti, and E. Gregori, IEEE 802.11 Wireless LAN: Capacity
Analysis and Protocol Enhancement, in INFOCOM, 1998.
Nevertheless, the performance anomaly analyzed in this [3] C. Koksal et al., An analysis of short-term fairness in wireless media
paper should be taken into account when we consider the access protocols, in Proceedings of ACM SIGMETRICS, 2000.

842
 12000 12000
Rate Bali Rate Bali

10000 10000
Throughput (Kbits/s)

Throughput (Kbits/s)
8000 8000

6000 6000

4000 4000

2000 2000

0 0
0 60 120 180 240 300 360 420 480 0 60 120 180 240 300 360 420 480
Time (secs) Time (secs)

Fig. 12. Evolution of the Balis bit rate in time Fig. 14. Evolution of the Balis bit rate in time

pairs that can be formed in the set of all hosts:

6000
Bali N 1 2
Marie Ps = N (N 1) =
5000 N
2
Throughput (Kbits/s)

. Thus, the average time of channel occupancy by the frames

4000 subject to a collision is
2 2
3000 tjam = Ts + (1 ) Tf .
N N
B. Estimate of tjam in the TCP case
2000
In the case of TCP traffic, the Access Point sends TCP
1000 ACKs at a variable rate that depends on the rate of the host
which is the destination of the ACK. For N 1 fast hosts and
0
a slow one, the Access Point sends (N 1)/2 more packets
0 60 120 180 240 300 360 420 480 than any other host.
Time (secs) Collisions may occur either between TCP data segments and
Fig. 13. Measured throughput in Mb/s for two hosts (one mobile), UDP
TCP ACKs, or between two TCP data segments. Each frame
traffic may be sent at the lower rate r or the higher rate R. In the
case of a collision between a TCP data segment and a TCP
ACK, the collision lasts the time needed for the transmission
[4] Netperf: a Network Performance Benchmark, 15 Feb. 1995, of the data segment. The number of host pairs that can be
hewlett-Packard, Information Networks Division. [Online]. Available: formed without the Access Point is (N 1)(N 2)/2 and
http://www.netperf.org/netperf/training/netperf.ps
[5] B. P. Crow et al., Investigation of the IEEE 802.11 Medium Access the number of hosts pairs that include it is N 1. As it sends
Control (MAC), in Proceedings of INFOCOM 1997, 1997. (N 1)/2 more packets, it contributes to the collision count
[6] M. Natkaniec et al., Analysis of Backoff mechanism in IEEE 802.11 for (N 1)2 /2.
standard, in Proceedings of ISCC 2000, 2000.
There are N 2 pairs that may include the host transmitting
[7] G. Bianchi, Performance Analysis of the IEEE 802.11 Distributed
Coordination Function, JSAC Wireless series, vol. 18, no. 3, 2000. at the lower rate. Only one pair includes the Access Point
and the slow host, but it contributes to the collision count for
A PPENDIX (N 1)/2.
Finally the probability Ps of having a packet sent at the
A. Estimate of tjam in the UDP case lower rate r involved in the collision is:
3 N 5
We assume the configuration described earlier with N 1 Ps = .
(N 1)(2 N 3)
fast hosts and a slow one. The duration of a collision depends
on the type of hosts involved in the collision. The probability And the average time of channel occupancy by frames subject
Ps of having a packet sent at the lower rate r involved in the to a collision is:
collision can be computed as the ratio between the number of tjam = Ps Ts + (1 Ps ) Tf .
host pairs that contain the slow host and the total number of

843