QoS Provided by the IEEE 802.

11 Wireless LAN to Advanced Data

Applications: a Simulation Analysis
G. Anastasi, E. De Stefano*, L. Lenzini
University of Pisa, Dept. of Information Engineering
Via Diotisalvi 2 - 56126 Pisa, Italy
email: [email protected]; [email protected]
* Graduate of the University of Pisa
Dept. of Information Engineering
Via Diotisalvi 2 - 56126 Pisa, Italy
Abstract
IEEE 802.11 is a Media Access Control (MAC) protocol which has been standardized by
IEEE for Wireless Local Area Networks (WLAN). The IEEE 802.11 MAC protocol
offers two types of services to its users: synchronous and asynchronous. This paper
thoroughly analyzes, by simulation, the asynchronous part alone, when each station feeds
the MAC protocol with data traffic patterns (in the following referred to as advanced data
traffic) that are very close in shape to those generated by a terminal accessing, for
example, the WWW service. Although the analysis shows that the performance of the
IEEE 802.11 with advanced traffic is considerably worse than with Poissonian traffic, we
can conclude that IEEE 802.11 still performs satisfactorily. Furthermore, our analysis is
broadened to include higher medium capacities up to 10 Mbit/sec. This part of the
analysis shows that the IEEE 802.11 MAC protocol is not adequate to work at speeds
planned for the upcoming ATM Wireless LAN.
1.0 Introduction
The IEEE 802.11 is a standard for a WLAN covering both Physical and MAC layers
([IEEE95], [LaMa95]). A key issue of this standard is that a mobile host is able to
communicate with any other mobile or wired host in a transparent manner. In other
words, an IEEE 802.11 WLAN appears to layers above the MAC layer as any other IEEE
802.X LAN (e.g. Ethernet or Token Ring). This means, in particular, that the mobility
aspects are handled at the MAC level or below.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 2 -
A performance analysis for the IEEE 802.11 WLAN is reported in [Chha96] where the
authors take into account the decentralized nature of communication between stations, the
possibility of capture, and the presence of hidden stations. They also study the impact of
spatial characteristics on system performances. Both parts (synchronous and
asynchronous) of the IEEE 802.11 MAC protocol have been analysed in [Viss95]. A
comparison of the IEEE 802.11 MAC protocol with HIPERLAN [ETSI95], the ETSI
proposal for wireless LANs, can be found in [Wein97].
The aim of this paper is to further advance the previous analysis on the asynchronous part
of the IEEE 802.11 standard. In fact, since we expect that relevant data applications
serviced by WLANs are very likely to resemble today's WWW applications, in our
performance analysis we use data traffic models which are very similar to those recently
proposed in the literature ([Deng96], [Crov95]) for these type of applications (i.e.,
WWW applications).
Furthermore, since discussions at an international level for a MAC protocol of an ATM
Wireless LAN have just began, we also analyse whether or not the IEEE 802.11 MAC
protocol is suitable for operating at higher channel speeds.
Although data integrity is a key requirement for data transmission, we assume an error
free radio channel despite the fact that we are dealing with a very unreliable environment.
This is because in the present paper we are focusing on multiple access aspects of the
IEEE 802.11 MAC protocol as measured by the average access delay experienced by
messages transmitted over data connections.
The paper is organized as follows. Section 2.0 describes the IEEE802.11 MAC
Protocol. Section 3.0 describes the simulation environment while results are reported in
Section 4.0. Section 5.0 concludes the paper.
2.0 IEEE 802.11 Wireless Network Standard
In this section we only report those aspects of the standard which are relevant for our
analysis. Specifically, Section 2.1 introduces the terminology and topology adopted for
an IEEE 802.11 WLAN, whereas Section 2.2 describes the MAC Protocol. See
[IEEE95] for details.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 3 -
2.1 Network Topology
An IEEE 802.11 WLAN generally consists of Basic Service Sets (BSSs) which are
interconnected by a Distribution System (DS) to form an Extended Service Set (ESS) as
shown in Figure 1.
BSS 1
BSS 2
802.x LAN
Distribution System
Portal
Access Point
Access Point
Figure 1: An IEEE 802.11 Network Topology
Each BSS consists of a group of wireless stations which execute a Distribution Function
(DF) to regulate the exclusive access to the shared wireless medium. Since the wireless
medium is broadcast, each station can transmit directly to any other station in the same
BSS. On the other hand, to transmit to stations belonging to a different BSS, stations
pass through an Access Point (AP) which is an interworking unit implementing both the
IEEE 802.11 and the DS MAC protocols.
However, before a station can access the wireless medium it needs to be associated with
an Access Point. A station can be associated with only one Access Point at a given time.
The Distribution System supports mobility by providing the necessary services for
handling destination mapping and the integration of BSSs in a manner that is transparent
to stations. This means that hosts (either wireless or wired) do not need to know the
physical location of other hosts for communications.
2.2 MAC Protocol
The IEEE 802.11 MAC protocol provides two service types: asynchronous, and
synchronous (or, better, contention free). These types of services can be provided on top
of a variety of physical layers and for different data rates. The asynchronous type of
service is always available whereas the contention free is optional.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 4 -
The asynchronous type of service is provided by the Distributed Coordination Function
(DCF) which implements the basic access method of the IEEE 802.11 MAC protocol and
is also known as the Carrier Sense Multiple Access with Collision Avoidance
(CSMA/CA) protocol.
The contention free service is provided by the Point Coordination Function (PCF) which
basically implements a polling access method. The PCF uses a Point Coordinator,
usually the Access Point, who cyclically polls stations giving them the opportunity to
transmit. Unlike the DCF, the presence of PCF is not mandatory. Furthermore, the PCF
itself relies on the asynchronous service provided by the DCF.
Since in our paper we do not analyze the contention free access method, in the following
only the DCF will be described. Details about the PCF can be found in [IEEE95].
According to the DCF a station must sense the medium before initiating the transmission
of a packet. If the medium is sensed as being idle for a time interval greater than a
Distributed InterFrame Space (DIFS) then the station transmit the packets. Otherwise, the
transmission is deferred and the backoff process is started. Specifically, the station
computes a random time interval, the backoff interval, uniformly distributed between zero
and a maximum called Contention Window (CW). This backoff interval is then used to
initialize the backoff timer. This timer is decremented only when the medium is idle
whereas it is frozen when another station is transmitting. Specifically, each time the
medium becomes idle, the station waits for a DIFS and then periodically decrements the
backoff timer. The decrementing period is referred to as the slot-time which corresponds
to the maximum round-trip delay within the BSS and, hence, depends on the maximum
BSS coverage.
As soon as the backoff timer expires, the station is authorized to access the medium.
Obviously, a collision occurs if two or more stations start transmission simultaneously.
Unlike wired networks (e.g. with CSMA/CD), in a wireless environment collision
detection is not possible. Hence, a positive acknowledgement is used to notify the
sending station that the transmitted frame has been successfully received. The
transmission of the acknowledgement is initiated at a time interval equal to the Short
InterFrame Space (SIFS) after the end of the reception of the previous frame. Since the
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 5 -
SIFS is, by definition, less than the DIFS
1
the receiving station does not need to sense
the medium before transmitting the acknowledgement.
If the acknowledgement is not received the station assumes that the transmitted frame was
not successfully received and, hence, schedules a retransmission and enters the backoff
process again. However, to reduce the probability of collisions, after each unsuccessful
transmission attempt, the Contention Window is doubled until a predefined maximum
(CWmax) is reached.
After a (successful or unsuccessful) frame transmission, if the station still has frames
queued for transmission, it must execute a new backoff process.
In radio systems based on medium sensing, a phenomenon known as the hidden node
problem may occur. This problem arises when a station is able to successfully receive
frames from two different transmitters but the two transmitters cannot receive signals
from each other. In this case a transmitter may sense the medium as being idle even if the
other one is transmitting. This results in a collision at the receiving station.
To deal with the hidden node problem the IEEE 802.11 MAC protocol includes an
optional mechanism which is based on the exchange of two short control frames: a
Request To Send (RTS) frame which is sent by a potential transmitter to the receiver and
a Clear To Send (CTS) frame which is sent from the receiver in response to the received
RTS frame. If the CTS frame is not received within a predefined time interval the RTS
frame is retransmitted by executing the backoff algorithm described above. After a
successful exchange of the RTS and CTS frames the data frame can be sent by the
transmitter after waiting for a SIFS.
The RTS and CTS frames (as well as data and acknowledgement frames) include a
duration field that specifies the time interval necessary to completely transmit the data
frame and the related acknowledgement. This information is used by stations which can
hear either the transmitter or the receiver to update their Net Allocation Vector (NAV), a
timer which, unlike the backoff timer, is always decremented, i.e., irrespective of the
medium status. Since stations which can hear either the transmitter or the receiver refrain
from transmitting until their NAV has expired, the probability of a collision due to a
hidden node occurring is reduced. Of course, the drawback of using the RTS/CTS
mechanism is an increased overhead which may be significant for short data frames.
1
The DIFS is defined as DIFS=SIFS+2Slot-times.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 6 -
The effectiveness of the RTS/CTS mechanism depends upon the length of the packet to
be "protected". In fact, it seems reasonable that RTS/CTS mechanisms improves the
performances when packet lengths are large when compared to the length of the RTS
control packet. Consequently there should be a threshold, the RTS threshold, under
which the mechanism is disabled and above which the mechanism is enabled.
3.0 Simulation Environment
The main difficulty for the analysis (at least) of the asynchronous part of the IEEE
802.11 MAC protocol via analytical models is the high degree of complexity and
interdependence of the various processes that are involved in the protocol operation.
Therefore, to evaluate the performances of the protocol we designed an ad hoc simulator
and implemented it in C++ language.
3.1 Traffic Characterization
Generally, when modelling data traffic, packet arrival processes are often assumed to be
Poissonian. However, this model does not capture any correlation between consecutive
message arrivals which is, instead, exhibited by experimental data ([Guse91] , [Lela94]).
To recover from this problem MMPP data traffic models have been proposed ([Guse91],
[Heff86]). One of the characteristics of this class of processes is that the sojourn time in
each state of the underline Markov chain is exponentially distributed. However, recent
teletraffic studies have shown that, experimental data related to WWW applications can be
satisfactorily modeled by ON/OFF processes where the ON and OFF time-length
distributions are heavily tailed (e.g., Weibull, Pareto)
2
.
Furthermore, in commonly used data applications (WWW, FTP, NNTP, Telnet, etc.),
messages vary in size with a distribution which deviates considerably from the simple
exponential distribution. Many studies have shown that the message size distribution is
bimodal (see, for example, [Guse91]) as shown by the probability mass function (pmf)
reported in Figure 2 which was estimated by a real trace [Bell].
2
As underlined in [Will95], by aggregating a large number of such ON/OFF sources, with infinite
variance distributions for the ON and OFF time durations, the resulting traffic process exhibits self-
similar properties.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 7 -
0
0.1
0.2
0.3
0.4
0.5
0
9
6
1
9
2
2
8
8
3
8
4
4
8
0
5
7
6
6
7
2
7
6
8
8
6
4
9
6
0
1
0
5
6
1
1
5
2
1
2
4
8
1
3
4
4
1
4
4
0
1
5
3
6
1
6
3
2
1
7
2
8
1
8
2
4
P
r
o
b
a
b
i
l
i
t
y

M
a
s
s

F
u
n
c
t
i
o
n
Packet Length (Bytes)
Figure 2: Packet size probability mass function.
For the purpose of our simulation analysis we use the following data source models:
1. Poisson ;
2. ON/OFF where the ON and OFF periods are distributed
exponentially (MMPP)
according to Weibull
In most of the analysis, for reasons explained in Section 4.0, arrivals within ON periods
are distributed exponentially.
It might be useful to recall that the Weibull distribution of a random variable X is
P X x F x e
x
{ } ( )

_
,

x 0
[1]
where
> 0
and
> 0
are real, and are called shape and scale parameters, respectively.
Furthermore, the average
E X [ ]
and variance

X
2
are
E X [ ]

_
,

1
[2]

X
2
2
2
2
2 1 1

_
,

_
,

1
]
1

'

[3]
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 8 -
where
z t e dt
z t
( )

1
0
.
From [1] it is easy to check that when
1
the Weibull distribution is reduced to an
exponential distribution.
3.2 Performance Measures
In this section we introduce the performance measures (or indices) used to characterize
the quality of service (QoS) a station attached to an IEEE 802.11 can achieve. The indices
defined in our analysis to specify the QoS are a widely accepted, minimum set of
performance measures used to characterize the performance of any computer network.
Before introducing the performance measures we observe that messages are queued in a
station local queue as soon as they arrive. Therefore, except when the station is empty, a
message will experience some delay in the local queue before contending for the channel
according to the IEEE 802.11 MAC protocol. With this in mind, the performance
measures we use in our analysis are:
- average queuing delay: time elapsed from the time a message joins the local queue up
until it reaches the head of the local queue itself: i.e., it starts contending for the
channel;
- average MAC delay: the average delay experienced by a message from the time it
reaches the head of the local queue up until the beginning of its successfull
transmission. The rationale behind this choice is that the average MAC delay measures
the delay caused by the MAC protocol, and thus measures the interference between a
user on one station and the users on the other stations. Furthermore, the average MAC
delay does not include the average queueing delay experienced by a message while
queued in the local node queue, i.e., it does not take into account the interference
between users on the same station;
- average access delay: sum of the mean MAC delay and the mean queueing delay in the
local queue;
- aggregate throughput: average number of bits successfully transmitted by all stations
per time unit.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 9 -
3.3 Operation Parameter Setting and Assessment Scenarios
The system parameters for our simulation environment are reported in Table 1. These
values are specified in the IEEE 802.11 standard [IEEE95]. Traffic related parameters are
reported in Table 2. Specifically, the station offered load value of 30 Kbit/sec was chosen
to cover a broad range of data services while the ON and OFF durations, although
different from the values in [Deng96], allow reasonable execution times for simulation
experiments without altering the stochastic nature of the arrival process.
System Parameter Parameter Value (sec)
Slot-time 50
SIFS 28
DIFS 128
(SIFS+2Slot-time)
Medium Capacity 1 Mbit/sec
Table 1: System parameter values
Traffic Parameter Parameter Value
Station Offered Load 30 Kbits/sec
Average ON duration 3.3 sec
Average OFF duration 22.8 sec
Table 2: Traffic parameter values
4.0 Simulation Results
In this section we report and comment on, in terms of MAC protocol mechanisms, the
simulation results we obtained with the arrival processes and operation parameter settings
previously specified. Furthermore, we report results which show the sensitivity of the
IEEE 802.11 MAC protocol to larger cell coverage and higher channel speeds.
4.1 Influence of the Arrival Process with RTS/CTS Disabled
In most of the curves reported below we compare Poisson, MMPP and ON/OFF
processes where ON/OFF durations are Weibull distributed with several coefficients of
variation (i.e., values). Furthermore, we also varied the interarrival packet time
distribution during ON periods. Specifically, we considered, in addition to the
exponential distribution, the Weibull and constant distributions. The results obtained were
substantially the same. For this reason, in most of the following experiments the
interarrival packet time distribution is taken as being exponential.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 10 -
Figure 3 shows the aggregate throughput achieved by the IEEE 802.11 vs the number of
data sources. This figure highlights that the throughput curves which refer to the various
arrival data processes are very close to each other.
0
100
200
300
400
500
600
0 5 10 15 20 25
Poisson
a=1.0
a=0.88
a=0.70
a=0.65
A
g
g
r
e
g
a
t
e

T
h
r
o
u
g
h
p
u
t

(
k
b
i
t
/
s
e
c
)
Number of stations
(MMPP)
Figure 3: Aggregate throughput vs the number of data sources for different arrival
processes.
0
200
400
600
800
1000
0 5 10 15 20
Poisson
a=1.0
a=0.88
a=0.70
a=0.65
A
v
e
r
a
g
e

A
c
c
e
s
s

D
e
l
a
y

(
m
s
e
c
)
Number of stations
(MMPP)
Figure 4: Average access delay vs the number of data sources for different arrival
processes.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 11 -
The situation is very different from the average access delay standpoint shown in Figure
4. This figure shows that the more the ON/OFF duration distributions deviate from the
exponential one (i.e., the more heavily tailed the distribution is) the worse the average
access delay. In order to understand the reasons for this behaviour we measured both
components of the average access delay. Figure 5 reports the average MAC delay and the
average queuing delay versus the number of stations for several values of .
0
200
400
600
800
1000
0
10
20
30
40
50
60
70
80
4 6 8 10 12 14
a=0.88
a=070
a=0.65
a=0.88
a=0.70
a=0.65
A
v
e
r
a
g
e

Q
u
e
u
i
n
g

D
e
l
a
y

(
m
s
e
c
)
Number of stations
Queueing
delay
MAC
delay
A
v
e
r
a
g
e

M
A
C

D
e
l
a
y

(
m
s
e
c
)
Figure 5: Average MAC delay and average queuing delay versus the number of stations
for several values of .
From the above figure it clearly emerges that the average MAC delay remains constant in
the range of values we considered and this means that the collision avoidance
mechanism is not influenced by the burstiness of the arrival process. By contrast, the
average queueing delay increases significantly when the values decrease. This
behaviour is certainly due to the fact that when the ON periods are distributed according
to a heavy tail distribution the probability of the occurrence of long ON periods is not
negligible and hence the probability of having a long local queue is not negligible either.
The influence of the value on the system performances is now clear. In the following,
in order to achieve reasonable execution times for simulation experiments, we will always
use the exponential distritibution (i.e., 1 ) for the OFF and ON durations. In other
words, the arrival process will be assumed to be MMPP.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 12 -
4.2 Influence of the Arrival Process with RTS/CTS Enabled
The purpose of this section is to analyze whether or not there exists an optimal RTS
threshold for the type of traffic we consider. Figure 6 shows a comparison between the
aggregate throughputs achieved by the IEEE 802.11 vs the number of data sources in the
cases in which the RTS/CTS mechanism is disabled and enabled and, in the latter case,
for different RTS threshold values. Specifically, the values selected are 0 (which means
RTS/CTS mechanism always enabled), 150 bytes and 280 bytes. This choice can be
understood by making reference to the left bump in the pmf reported in Figure 2.
Obviously, 150 bytes falls in the middle and 280 on the right side of the above mentioned
bump.
0
200
400
600
800
1000
0 5 10 15 20 25 30
Thresh. = 0 bytes
Thresh. = 150 bytes
Thresh. = 280 bytes
RTS/CTS Disabled
A
g
g
r
e
g
a
t
e

T
h
r
o
u
g
h
p
u
t

(
K
b
i
t
/
s
e
c
)
Number of stations
Figure 6: Aggregate throughputs when the RTS/CTS mechanism is enabled (for different
threshold values) and disabled.
As shown in Figure 7, the RTS/CTS protocol mechanism significantly influences the
average access delay for any RTS threshold value. This figure highlights two important
aspects:
- the RTS/CTS mechanism improves the average access delay with respect to the case in
which RTS/CTS is disabled;
- when the RTS/CTS mechanism is enabled, the influence of the RTS threshold value on
the average access delay is almost negligible although it can be observed that the
average access delay decreases when the RTS threshold value decreases.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 13 -
The latter point can be easily understood by taking into consideration that, due to the
shape of the packet size pmf (see Figure 2):
- by lowering the RTS threshold value, the number of packets affected by the RTS/CTS
mechanism increases, however
- the improvement in the RTS/CTS mechanism is very strong for packets belonging to the
right bump (packets with very large packet sizes) while it is moderate or even low for
packets falling in the left bump (packets with small to medium packet sizes). Hence,
the improvement in the access delay performance of the RTS/CTS mechanism is not
significantly affected by the positioning of the threshold within the left bump.
0
200
400
600
800
1000
0 5 10 15 20
Thresh. = 0 bytes
Thresh. = 150 bytes
Thresh. = 280 bytes
RTS/CTS Disabled
A
v
e
r
a
g
e

A
c
c
e
s
s

D
e
l
a
y

(
m
s
e
c
)
Number of stations
Figure 7: Average access delay when the RTS/CTS mechanism is enabled (for different
threshold values) and disabled.
To further deepen the influence of the RTS threshold value on the average access delay
performance we investigated other packet length distributions. Specifically, for the
uniform and exponential distributions with the same average as the distribution reported
in Figure 2, we observed the same behaviour: i.e., the threshold value does not affect
significantly the access delay protocol performance.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 14 -
4.3 Influence of the Slot-time Duration
As highlighted in Section 2.2, the slot duration depends upon the maximum size of the
cell coverage. Figures 8 and 9 report the throughput and average access delay vs the
number of stations for several cell sizes expressed in terms of slot durations.
0
100
200
300
400
500
600
0 5 10 15 20
50 sec
100 sec
200 sec
300 sec
A
g
g
r
e
g
a
t
e

T
h
r
o
u
g
h
p
u
t

(
K
b
i
t
/
s
e
c
)
Number of stations
Figure 8: Influence of the slot-time duration on the aggregate throughput.
0
200
400
600
800
1000
1200
1400
0 5 10 15 20
50 sec
100 sec
200 sec
300 sec
A
v
e
r
a
g
e

A
c
c
e
s
s

D
e
l
a
y

(
m
s
e
c
)
Number of stations
Figure 9: Influence of the slot-time duration on the average access delay.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 15 -
Figure 8 shows that the maximum achievable throughput decreases when the slot
duration increases. Furthermore, from Figure 9 it follows that the average access delay,
for a given number of stations, increases when the slot duration increases. This can be
explained by taking into consideration that an increase in the slot duration results in: i) an
increase in the DIFS and hence in the average backoff period; and ii) an increase in the
number of collisions, as shown in Table 3.
Slot-time Duration
Number
of
Stations
50 sec 100 sec 200 sec 300 sec
13 0.19 0.20 0.23 0.26
16 0.28 0.31 0.35 0.50
19 0.56 0.81 1.05 1.12
Table 3: Mean number of collisions versus slot-time duration for three different numbers
of stations.
The latter effect can be explained by considering that the vulnerable period (see Figure 10)
of the collision avoidance algorithm implemented by the IEEE 802.11 MAC protocol has
a duration equal to the slot time. This can be understood by analyzing Figure 10 which
shows a scenario in which the backoff time of a station (called reference station) expires
at time t
0
. Due to the propagation delay, any other stations for which the backoff expires
in between half of a slot before and after t
0
(stations 1 and 2 respectively - see Figure 10)
will collide with the reference station.
Vulnerable
period
(1 slot-time)
backoff
backoff1
backoff2
Reference station
Station 1
Station 2
t0
t2
t1
Figure 10:Vulnerable period.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 16 -
4.4 Analysis at Higher Speeds
As attention is now turning towards wireless LANs which can support several tens of
megabit per second, we broadened our analysis to verify whether or not the IEEE 802.11
MAC protocol is suitable for managing these rates efficiently.
We estimated the aggregate throughput versus the number of stations for channel speeds
of 1, 2, 5, and 10 Mbit/sec. The results obtained show that the number of stations for
which the maximum aggregate throughput occurs does not increase proportionally to the
channel speed. This is due to the fact that when the channel speed increases, the message
transmission time decreases proportionally. However, since the slot time, and hence the
backoff time, remains unchanged with the increase of the channel speed, the portion of
time during which the channel remains unused due to the backoff algorithm is the same at
any speed.
0.25
0.3
0.35
0.4
0.45
0.5
0.55
0 2 4 6 8 10
P
r
o
t
o
c
o
l

C
a
p
a
c
i
t
y
Medium Capacity (Mbit/sec)
Figure 11: Protocol capacity vs. different speed rates
Figure 11 plots the IEEE 802.11 MAC protocol capacity (i.e., the maximum fraction of
channel bandwidth used by successfully transmitted messages over all possible offered
loads) for 1, 2, 5, and 10 Mbit/sec. As can be seen, the protocol capacity decreases when
the channel speed increases. This implies that: i) the IEEE 802.11 is not adequate to
support the channel speeds planned for the future generation of wireless ATM LANs; and
ii) the protocol capacity depends upon the ratio (denoted by a in the literature - see
[Cont96]) between the packet transmission speed and the medium propagation time. The
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 17 -
latter property should be expected since the IEEE 802.11 MAC protocol belongs to the
class of random access MAC protocols which exhibit a similar dependency.
5.0 Summary and Conclusions
In this paper we have performed a simulation analysis of the IEEE 802.11 MAC protocol
proposed for wireless LANs. This analysis is based on realistic traffic models which
are in line with those derived for modern data network applications (e.g., WWW).
From the above analysis we can conclude that when the number of stations is in the order
of 15, the IEEE 802.11 behavior is satisfactory. Under this condition, the IEEE 802.11
resource-sharing distributed algorithm results in an access delay experienced by each
station in the order of one second (in our experiments) when the RTS/CTS is disabled,
and approximately one third of second when the RTS/CTS mechanism is enabled. This is
a boundary which seems to be acceptable for today's data services.
Furthermore, at least for the type of traffic we consider, results obtained suggest using an
RTS threshold equal to zero; i.e., to enable the RTS/CTS protocol mechanism for any
message length.
Finally, the IEEE 802.11 MAC protocol is not suitable for managing data traffic at
channel speeds higher than a few megabit/sec.
References
[Bell] Trace retrived via anonymous ftp from bellcore.flash.com, file
/pub/lan_traffic/pAug.TL. No longer available.
[Chha96] H. S. Chhaya, S. Gupta, Performance of Asynchronous Data Transfer
Methods of IEEE 802.11 MAC Protocol, IEEE Personal
Communications, October 1996, Vol. 3, No 5.
[Cont96] M. Conti, E. Gregori, L. Lenzini, Metropolitan Area Networks,
Springer 1996.
[Crov95] M. E. Crovella, A. Bestavros, Explaining World Wide Web Traffic Self-
Similarity, Computer Science Dept., Boston University, Technical Report
TR95-015, August 29, 1995.
Workshop on Nomadic Computing
Geneva (CH) - April 5, 1997
- 18 -
[Deng96] S. Deng, Empirical Model of WWW Document Arrivals at Access Link,
Proceedings of ICC 96.
[Heff86] H. Heffes. D. M. Lucantoni, A Markov Modulated Characterization of
Packetized Voice and Data Traffic and Related Statistical Multiplexer
Performance, IEEE JSAC, vol. SAC-4, No. 6, September 1986.
[ETSI95] ETSI, Hiperlan Functional Specification, ETSI Draft Standard, July 1995.
[IEEE95] P802.11/D2.1, Draft Standard IEEE 802.11, Wireless LAN Medium
Access Control (MAC) and Physical Layer (PHY) Specifications, 4
December 1995.
[Guse91] R. Gusella, Characterizing the Variability of Arrival Processes with
Indexes of Dispersion, IEEE JSAC, vol. SAC-9, No. 2, February 1991.
[LaMa95] R. O. LaMaire, A. Krishna, J. Panian, and P. Bhagwat, Wireless LANs
and Mobile Networking: Standards and Future Directions, IBM
Corporation, December 22, 1995.
[Lela94] W. E. Leland, M. S. Taqqu, W. Willinger, D. V. Wilson, "On the Self-
Similar Nature of Ethernet Traffic", IEEE/ACM Transactions on
Networking, Vol. 2, No 1, pp. 1-15, February 1994.
[Viss95] M. A. Visser, M. El Zarki, Voice and Data transmission over an 802.11
Wireless network, Proceedings of PIMRC 95.
[Wein97] J. Weinmiller, M. Schlager, A. Festag, A. Wolisz, Performance Study of
Access Control in Wirelles LANs - IEEE 802.11 DFWMAC and ETSI
RES 10 HIPERLAN, To appear on ACM/Baltzer Wireless Network,
1997.
[Will95] W. Willinger, M. S. Taqqu, R. Sherman, D. V. Wilson, Self-Similarity
through High Variability Statistical Analysis of Ethernet LAN Traffic at the
Source Level, Proceedings of ACM SIGCOMM 95, pp. 100-113, 1995.