Sunteți pe pagina 1din 5

1 Performance Analysis of the IEEE 802.11 Wireless LAN Standard C.

Sweet

Performance Analysis of the IEEE 802.11 Wireless LAN Standard 1


Craig Sweet and Deepinder Sidhu

Maryland Center for Telecommunications Research


Department of Computer Science and Electrical Engineering
University of Maryland Baltimore County
1000 Hilltop Circle
Baltimore, MD 21250

{sweet, sidhu}@cs.umbc.edu

1
This research was supported in part by the Department of Defense at the Maryland Center for Telecommunications Research,
University of Maryland Baltimore County. The views and conclusions contained in this document are those of the authors and
should not be interpreted as representing the official policies, either expressed or implied, of the Department of Defense or the
U.S. Government.
2 Performance Analysis of the IEEE 802.11 Wireless LAN Standard C. Sweet

1. Abstract in this new direction in and can significantly alter its In this analysis, we performed
networking. results. four experiments measuring
IEEE 802.11 is a relatively new various aspects of the MAC
standard for communication in a 3. Modeling and Simulation Description of Simulation protocol. Each of these
wireless LAN. Its need arose Model experiments was conducted at
from the many differences In our experiments, the goal several transmission speeds. 1
between traditional wired and was to explore the efficiency of and 2 Mbit/s were selected
To perform this analysis, we
wireless LANs and the the MAC protocol under ideal because they are explicitly
constructed a discrete-event
increased need for conditions. While many of supported in the specification.
simulation of the MAC portion
interoperability among different these conditions may be 10 Mbit/s was selected to
of the IEEE 802.11 protocol. A
vendors. To date, detailed unrealistic, the end result is provide a comparison at
complete description of
performance measures for this useful in telling us the highest traditional LAN speeds. The
simulation techniques can be
CSMA/CA protocol are not performance that can be results of these experiments are
found in [BAN84]. For all
known. We describe the results expected from the protocol. the topic of this section.
experiments, each station is
of our Discrete-Event This section describes some of Current research is aimed at
assumed to have a one MAC
Simulation of the Distributed the assumptions and limitations providing 802.11 operation at
Service Data Unit (MSDU)
Coordination Function (DCF) assumed in our system. Also, 10 and 20 Mbit/s.
buffer. An MSDU is the basic
within the MAC sublayer. We the simulation model and
unit delivered between two
model an ideal LAN and computation variables are Experiment 1: Variable Load
compatible MAC sub-layers.
describe the best case described.
For uniformity all MSDUs
performance. Our results show
transmitted are of equal size. In our first experiment we
the relationship between the Assumptions Initially, each station is given wanted to see what effect the
protocol options and total
one MSDU to transmit. Upon total load on the system played
system throughput.
All stations are assumed to be completing the transmission on performance. This
using a Direct Sequence Spread attempt, another MSDU is experiment is similar to one
2. Introduction
Spectrum (DSSS) radio. The assigned for transmission after found in [GON87]. Figure 1
operation of Frequency Hopping some exponentially distributed shows the variation of total
Over the last several years, we
Spread Spectrum (FHSS) and inter-arrival time. In this throughput with total offered
have witnessed widespread
Infrared (IR) radios had too manner, changing the mean load G for various message
deployment of Wireless LANs
much of an impact on a given inter-arrival time between sizes P at 1 Mbit/s. In this
in virtually every industry.
transmission to study the MSDUs can be used to alter experiment, the fragmentation
Until recently, there has been no
aspects of the protocol itself. system load. threshold has been set to 2346
agreed upon standard by which
Additionally, it is assumed that bytes and the RTS threshold has
wireless stations communicate.
there are no power Offered Load Computation been set to 3000 bytes.
This lack of standardization
considerations for either the
usually results in decreased
radios or the wireless stations We can see that with an offered
interoperability. The Industry Upon transmitting a message,
that could interfere with the load of about 80% or less
for Electrical and Electronics the station generates the next
operation of the protocol. virtually no collisions occur and
Engineers (IEEE) has been message with an inter-arrival
throughput and load are
working with leaders from time exponentially distributed
A significant aspect of any approximately equal. Once the
industry to develop a standard
transmission protocol is how it with mean . Additionally, system load increases beyond
to which wireless stations from each station is sending the same
handles transmission errors. In 90-100% we see the impact of
different vendors can size packets, in bytes P, for the
order to focus on the core MAC collisions. As can be expected,
communicate. In 1997, the duration of a run. The offered
protocol, we assumed error-free greater throughput is achieved
IEEE finally ratified their load of station i, Gi, is defined
channels. Additionally, all via a greater packet size. Due
standard 802.11, the Physical as in [GON87] to be the
stations have unobstructed to the overhead present in the
and MAC specification for throughput of station i if the
access to all other stations and protocol, acceptable throughput
Wireless LANs [IEE97]. network had infinite capacity,
thus can hear all transmissions. was not seen with packet sizes
i.e., below 2000 bytes.
Since traditional Ethernet has
To minimize complexity, we
been in existence for quite some
time, much research has been
chose to model our wireless Gi = Tp / I
LAN as an ad-hoc network, also where Tp = P/C and C is the
done studying its attributes
known as an Independent Basic transmission speed in Mbit/s.
under various conditions
Service Set (IBSS). This is the The total offered load can thus
[BUX81, GON83, and GON87].
simplest type of wireless LAN be computed to be
A detailed study of the Carrier
defined in the standard. There N
Sense Multiple Access with
Collision Detection scheme
is no Access Point and therefore
no tie to a wired LAN. G i
used in Ethernet can be found in i 1
[TOB80].
Finally, we decided to only
4. Performance Analysis
model the DCF portion of the
Since wireless networking is a
protocol. The DCF forms the
recent development, not much is
heart of the MAC protocol. The
known about how the protocol
PCF has been omitted in this
performs. The goal of this
simulation, as it is an optional
research is to uncover some of
portion of the MAC protocol
the hidden performance issues
that works on top of the DCF
3 Performance Analysis of the IEEE 802.11 Wireless LAN Standard C. Sweet

As in the previous experiment,


Net Throughput, %

100
P=4500
P=3200

No RTS RTS
80 100
P=2800
90
P=2347
80
P=512
70
60
60
P=128
50

40

40 30

20

10

0
20 2 4 8 16 32 64 128 256 512 1024
Number of Stations
Fig. 2. Throughput vs. Number of stations with packet size above
and below RTS Threshold (1 M bit/s)
0
10 20 40 100 200 400 800 1000
Results are shown both with we saw similar results in our 2
and without the RTS and 10 Mbit/s experiments. As
Total Offered Load, %
mechanism implemented. For the speed of the medium
Fig. 1. Throughput vs. Offered Load at 1 Mbit/s, 32 stations, all runs, the message size was increased there was still the
Parameter, P.
set to 3000 bytes and the same pattern between RTS and
4500 96.11 fragmentation threshold was set No RTS results. We can see
Packet sizes above and below 2800 76.07 to 2346 bytes. that higher transmission speeds
the fragmentation threshold did 2347 73.08 yielded lower average
not yield much difference. Even Without RTS enabled, we can throughput results. Table II
4500 91.80
then, it all but disappeared with10 see that the maximum summarizes some of the results
2800 73.17
loads in excess of 200%. While throughput reached was approx. from these experiments with
2347 68.87
increasing the number of 82% with 16 participating and without RTS enabled.
packets per message produces stations. In fact, with few
more overhead, it also reduces Our results were similar for stations (below 16), we see that Table II
the collision probability. transmission speeds of 2 and 10 there is not much difference Simulation
in Results at 100% Offered Load with Variabl
Mbit/s. Table I summarizes performance with and without Number of Stations
In this example, the RTS some of these results. What we RTS enabled.
threshold played a crucial role saw was that as the Throughput Throug
in the performance of the transmission speed increased, As more stations are added to # of Stations with RTS without
protocol. The throughput the throughput dropped. This the simulation the probability 16 82.59 81.9
peaked out at approx. 80% for can be attributed to the fact that that two or more stations will
the inter-frame spaces are 128 92.38 76.2
all packet sizes below 3000 calculate the same backoff
bytes. For packet sizes above independent of transmission window is increased. Thus, the 1024 94.81 46.1
the RTS threshold, noticeable speed. At higher speeds, since chance for collision increases. 16 82.83 82.9
performance gains were seen it takes less time to send the This can be seen by the large
same packet, an Inter-Frame 128 93.04 73.2
and throughput peaked at 96%. differences between the RTS
Space (IFS) of 50 s has more and No-RTS runs with higher 1024 94.07 53.9
The RTS threshold acts as a of an impact than at lower station counts, above 64. 16 80.34 78.0
medium reservation mechanism. speeds. 128 88.6 70.0
Collisions, and subsequent Since IEEE 802.11 uses
1024 89.95 57.8
retransmissions, can occur on Experiment 2: Variable CSMA/CA, collisions are
the smaller RTS frames but not Stations expensive. The transmitting
normally on the longer data station must continue to
frames. The result is a better transmit the entire message and Experiment 3: Variable
In our second experiment, our
utilization of the bandwidth. wait a minimum amount of time Fragmentation
goal was to determine how
many stations would overload a before determining that the
Table I wireless network. Certainly the transmission was in error. With In our third experiment, our
Simulation Results at 200% Offered Load for Various performance characteristics for RTS enabled, the collisions goal was to determine what
Packet Sizes and Transmission Speeds 10 stations would be different occur on smaller RTS frames, effect the fragment size played
than for 20 stations, all allowing for a quicker turn- on system performance. The
contending for access to the around time. We can see that simulation was run with 32
Mbit/s Packet Size Throughputmedium.
% Figure 2 shows the with RTS enabled, the system stations at 200% load with
4500 96.61 effect on throughput with an stabilized to approx. 92% or varying fragmentation
1 2800 76.54 increasing number of stations higher with 128 or more thresholds. Each message sent
2347 71.52 and a constant Offered Load of stations. was 3000 bytes long.
100%. Therefore, the fragmentation
4 Performance Analysis of the IEEE 802.11 Wireless LAN Standard C. Sweet

threshold merely determined collision requires 250 79.61


how many fragments the 3000 retransmission of a much larger 1250 93.46 When a station transmits a
byte messages were broken up fragment. Since 802.11 does 3000 94.96 message, it waits only a finite
into. not have a collision detection amount of time for the response.
mechanism the entire fragment 250 78.44 If this response does not arrive
Intuitively, advantages can be must be transmitted before 1250 92.68 in time, it will retransmit the
gained by both increasing and success or failure of that 3000 94.27 message. This timer begins
decreasing the fragmentation fragment can be determined. immediately after the sender
threshold. Smaller thresholds 250 70.47 finishes transmitting the
limit the loss of performance This experiment has also shown 1250 88.79 message. If the receiver is
due to retransmissions but come that, in this specific case, little 3000 89.45 sufficiently far away from the
with an increase in overhead. improvement can be seen with sender, much of this time is
This is important because the fragments above 1000 bytes taken up by twice the delay
802.11 protocol has when the RTS mechanism is Experiment 4: Variable between the stations, once for
considerable overhead [IEE97]. used. While this may be true in Propagation Delay the message to reach the
On the other hand large this experiment, note that we recipient and once for the
fragmentation thresholds, while are assuming that all fragments response to arrive at the source.
limiting the overhead, become are transmitted error-free. This In our previous experiments, we
expensive in the event of a assumption will certainly not assumed a constant delay of 1 If the distance between two
collision. hold in a real-world case. In s between stations. This stations becomes too large, it
allowed us to measure the will be impossible for the
protocol performance without sender to hear the
respect to the interoperability in acknowledgement from the
a real-life situation. In our receiver. In this case, it
No RTS RTS fourth experiment, our goal was becomes increasingly difficult
1 00 to determine how far apart for messages to be received
stations can be from one correctly. The result is
another, in terms of propagation increased retransmissions and
90
delay, before system throughput decreased throughput.
degrades. In a real-world
80
wireless network, some stations
may be constantly moving while
others are stationary for periods
70 of time.

In this experiment, we set the


60

250 500 750 1 000 1 250 1 500 1 750 2000 2346 3000 fragmentation threshold to 2346
Fragmentation Threshold bytes and the message size to
Fig. 3 Throughput vs. Fragmentation Threshold with packet size above 3000 bytes. The system is run
and below RTS Threshold (1 M bit/s) at 100% Offered Load. Figure
fact, performance may decrease 4 shows the results of
Figure 3 shows the results of as the bit-error rate increases. increasing the propagation delay
this experiment run at 1 Mbit/s. The probability of each between any two wireless
As predicted, the RTS fragment being successfully stations operating at 1 Mbit/s.
mechanism does a great deal to delivered will decrease as the
improve the performance of this fragment size increases and We can see that, with the
aspect of the protocol. The results will most certainly current fragmentation threshold
reason can be attributed to the differ. and a 50 s IFS, throughput
reduction in collisions that it drops when the propagation
provides. At smaller thresholds, The results for 1, 2, and 10 delay between stations exceeds
there is little difference between Mbit/s experiments are 50 s.
the RTS and No RTS figures. summarized in Table III. We
There is nearly a balance can see that the same pattern is
between three factors: the exhibited regardless of the
overhead provided by the RTS transmission speed. As we have No RTS RTS
mechanism, the smaller seen in the previous 1 00

fragment sizes that are experiments, the constant inter- 90

retransmitted in the event of a frame space times effectively 80


collision, and the overhead reduce the system performance 70
provided by multiple smaller at higher speeds.
fragments. 60

Table III 50

It is not until the fragmentation


Simulation Results at 200% Offered Load with Variable 40

threshold increases that we seeFragmentation Threshold and Transmission Speed 30

the largest variation in 20


performance. As was expected, Frag. Throughput 10
with larger fragments comes a Threshold with RTS
0
decrease in performance. Each 2 4 8 16 32 64
Propagation Delay Between Stations
Fig. 4. Throughput vs. Propagation Delay with packet size above and below
RTS Threshold (1 M bit/s)
5 Performance Analysis of the IEEE 802.11 Wireless LAN Standard C. Sweet

overhead, offers considerable [GON87] T. A. Gonsalves, "Measured


improvement in most highly Ethernet," in Advances in Lo
loaded systems. Kummerle, K., Tobagi, F., an
No RTS RTS York: IEEE Press, 1987.
1 00 We found that the best [IEE97] IEEE Std 802.11-1997, “IEE
90 performance can only be Metropolitan Area Networks:
80
achieved in systems with Access Control (MAC) and P
70
relatively slow transmission Specification.”
60
speeds. Transmission speed [TOB80] F. A. Tobagi and V. B. Hunt,
and throughput were inversely carrier sense multiple access
50
proportional. This is due to the Comput. Networks, vol. 4, O
constant delays and timers used
40

30
in the protocol, which are not
20
altered as transmission speed
10
increases.
0
2 4 8 16 32 64

Propagation Delay B etween Stations Future Work


Fig. 5. Throughput vs. Propagation Delay with packet size above
and below RTS Threshold (2 M bit/s) Currently our research does not
take into account the
Unfortunately the problem only It a reasonable assumption that transmission errors that are
compounds itself as the there is a limit to the distance inherent in all forms of
transmission speed increases. that any two communicating communication. One area of
Figure 5 shows the same stations can be from one another research will be to incorporate a
experiment run at 2 Mbit/s. before system performance bit-error rate into the
Here we see that the same drop suffers. This limit is based simulation, based upon the
off in throughput occurs with upon the attributes of the transmission device, and see
stations only 32 s apart. The communication medium and the how the system performance is
reason is that the initial protocol. From this experiment affected.
transmission is shorter at the we can see that the transmission
higher speed, which forces the speed also plays a crucial role. Our system did not allow for a
station to begin its waiting subset of stations to be hidden
period earlier. Therefore, this 5. Conclusion from the others. We assumed
timer can expire with a shorter that all stations can hear all
propagation delay. While the experiments transmissions from all others.
described in this paper do not With this medium, stations can
As can be expected, the results reflect any real-life scenario, be obstructed from some other
are even worse for they are useful in determining stations in the network. This
transmissions at 10 Mbit/s. the maximum system would prevent them from
These results are shown in performance under a variety of reading all of the Network
figure 6. An interesting point in conditions. Our goal has been Allocation Vector (NAV) values
all three graphs is that the RTS to see what the maximum that are transmitted. Future
mechanism can do little to performance we can expect out research could take this into
improve this performance. This of the protocol is and what it account.
assures us that the loss in takes to reach it.
throughput is not attributed to The aim of our research was
collisions but rather to too much We see from our experiments focused on the DCF but
distance between stations. In that Ethernet speeds are completely ignored the optional
fact, the added overhead of the possible but only with the RTS PCF. It is quite possible that
RTS mechanism slightly mechanism that is built into the some of the inefficiencies found
reduces the performance once 802.11 MAC protocol. This in our experiments can be
this problem occurs. mechanism, while adding some overcome by the PCF. Our
current research aim is to
explore possible performance
gains by exploiting the PCF
functionality.
No RTS RTS
100
References
Net Throughput, %

90

80
[BAN84] Banks, J. and J. S. Carson, “Discrete-Event System
70
Simulation,” Prentice-Hall, Englewood, NJ, 1984.
60
[BUX81] W. Bux, "Local-area subnetworks: A performance
50
comparison," IEEE Trans. Commun
40
1465-1473, 1981.
30 [GON83] T. A. Gonsalves, "Performance characteristics of 2
20
Ethernets: An experimental study," ACM SIGCOMM
10 Symp. On Commun. Architectures and Protocols
0 Austin, TX, Mar. 1983, pp. 178-185.
2 4 8 16 32 64
Propagation Delay Betw een Stations
Fig. 6. Throughput vs. Propagation Delay w ith packet size above
and below RTS Threshold (10 Mbit/s)

S-ar putea să vă placă și