IEEE 802.

11 Packet Delay – A Finite Retry Limit

Analysis
P. Chatzimisios, A. C. Boucouvalas and [Link]*
Multimedia Communications Research Group,
School of DEC, Bournemouth University
Poole, UK
{pchatzimisios, tboucouv}@[Link], vitsas@[Link]

Abstract—The contribution of this paper is on the study of packet Send (RTS/CTS) reservation based scheme. This scheme uses
delays for the IEEE 802.11 wireless local area network DCF the small RTS/CTS packets to reserve the medium when large
MAC protocol. A method is presented capable of taking into packets are transmitted in order to reduce the duration of a
account retransmission delays with or without retry limits. We collision and to deal with the hidden terminal problem [5].
present an analytical model based on a Markov chain which
In the literature, several papers have studied the
allows us to derive closed form expressions for the packet delays,
the probability of a packet being discarded when it reaches the performance of the IEEE 802.11 protocol using analytical
maximum retransmission limit and the average time to drop such modes [6] [7] or by means of simulation [8] [9]. Moreover,
a packet for the basic and RTS/CTS access mechanisms. The much research effort is focused on improving the performance
results presented are for standard protocol parameters versus the of the 802.11 MAC [10][11] and showed that performance
number of contention stations. Finally, the accuracy of the depends on both throughput and delay considerations.
analytical model is verified by simulations. Our work introduces a performance analysis of both basic
Keywords; wireless; IEEE 802.11; packet delay; retry limit; access and RTS/CTS access mechanisms. A throughput
mathematical analysis has been developed [12] that does not
I. INTRODUCTION consider packet dropping when the retry limit is reached. This
Wireless local area networking (WLAN) is a very dynamic throughput model is extended in [13] to include packet
field and a rapidly growing communication business. dropping. We have reported the packet delay in [14], using the
Technological and regulatory developments have allowed the Markov chain model of [12] which does not consider
issues of high prices, low data rates and licensing retransmission limit on collisions. In this paper we provide a
requirements to be addressed and the popularity of wireless new packet delay analysis, which extends the analysis in [14]
LANs has grown significantly over the past few years. With to include the effect of retry limits. More specifically, this
wireless networking, regardless of where end users are within delay analysis allows us to calculate the packet delay, the
range, they can have network connectivity and they are a packet drop probability and the packet drop time of the 802.11
mouse-click away from key information and applications. protocol. Additionally, our model is compared with a model
To deal with wireless user connectivity needs, various [12] that does not consider packet dropping which occurs
wireless communication standards have been developed when the number of packet transmission retries attains its
[1][2]. The IEEE 802.11 protocols are a significant limit. By including the packet retry limit, we consider that the
development; they are now a mature technology for WLANs model predicts the 802.11 packet delay in an accurate way.
and can offer high data rates through 802.11a and 802.11b Our model assumes that the network consists of n
standards, [3] [4]. The specifications are detailed and include contending stations transmitting in ideal conditions (no errors
both the Medium Access Control (MAC) and the Physical occur in the channel and no hidden stations exist). We also
Layer (PHY). The MAC incorporates two medium access consider that every station has always a packet available for
methods, Distributed Coordination Function (DCF) and Point transmission (saturated conditions). The key assumption in our
Coordination Function (PCF). DCF is an asynchronous data model is that the collision probability of a transmitted packet
transmission function, which is best suited to delay insensitive is constant and independent of the retransmissions that this
data. On the other hand, the optional Point Coordination packet has suffered in the past.
Function (PCF) is utilized, when time-bounded services are This paper is organized as follows: Section II introduces the
required according to the communications needs. DCF of the IEEE 802.11 MAC and focuses on the backoff
Under DCF, data packets are transferred via two schemes. procedure. Section III reviews the mathematical model. In
The default scheme is called the basic access mechanism, which Section IV, we extend the model and develop an analytical
transmits the data packet after deferring if the medium is busy. method to predict packet delay. Section V provides various
The 802.11 standard also provides an optional way of analysis results. These results allow us to determine how the
transmitting data packets, namely the Request To Send/Clear To protocol performance is affected by the retry limit. Finally,
section VI concludes the paper.
* Department of Information Technology,
Technological Educational Institution, Thessaloniki, Greece

GLOBECOM 2003 - 950 - 0-7803-7974-8/03/$17.00 © 2003 IEEE

 II. DISTRIBUTED COORDINATION FUNCTION (DCF) not receive an ACK, the data packet is assumed to have been
DCF is based on the Carrier Sense Multiple Access with lost and a retransmission is scheduled.
Collision Avoidance (CSMA/CA) technique and adopts a III. ANALYTICAL MODEL
slotted Binary Exponential Backoff (BEB) scheme to reduce
collisions due to stations transmitting simultaneously. Each Let b(t) and s(t) be the stochastic processes representing the
node with a packet to transmit first senses the medium to backoff timer and the backoff stage respectively for a given
ascertain whether it is in use. If the medium is sensed to be idle station at slot time t.
for a time interval greater than the Distributed Inter-Frame
Space (DIFS), the station proceeds with the packet 1 1 1 1 1
transmission. If the medium is sensed busy, the station defers 1-p
0,0 0,1 0,2 zzz 0,W0-2 0,W0-1

transmission and initializes its random backoff interval. This

backoff timer is decremented when the medium is idle and is p/W1

frozen when the medium is sensed busy. After a busy period zzz zzz zzz zzz zzz zzz

the backoff resumes only after the medium has been idle for 1 1 1 1 1
longer than DIFS. 1-p
i,0 i,1 i,2 zzz i,Wi-2 i,Wi-1

Every station maintains a station short retry count (SSRC)

as well as a long retry count (SLRC), both of which take an i+1,0 i+1,1 i+1,2 zzz
p/Wi+1
zzz i+1,Wi+1-2 i+1,Wi+1-1
initial value of zero for every new packet. The short retry count 1-p 1 1 1 1 1

indicates the maximum number of retransmission attempts of a zzz zzz zzz zzz zzz zzz zzz
RTS packet or of a data packet when RTS/CTS is not used.
The long retry count indicates the maximum number of
retransmission attempts of a data packet when RTS/CTS is p/Wm
zzz
m,0 m,1 m,2 zzz m,Wm-2 m,Wm-1
used. When either of these limits is reached, retry attempts 1 1 1 1 1 1

cease and the packet is discarded. We assume an error free Fig. 1 Markov chain model
channel, no hidden stations and packets are retransmitted only
when they encounter collisions. As a result, the long retry limit A discrete-time Markov chain showed in fig. 1 is used to
is not used in our analysis. model the bi-dimensional process {b(t), s(t)}. Let
The random number for the backoff timer is chosen in the bi,k = lim P{ s ( t ) = i , b (t ) = k } be the stationary distribution
t→∞
interval (0,CW-1), where CW is the contention window size.
of the Markov chain, where i∈[0,m] , k∈[0,Wi-1]. We have the
The value of CW depends on the number of failed
following relations:
transmissions of a packet. At the first transmission attempt
W=CWmin , which is the minimum contention window. After bi,0 = p. bi-1,0 , 0<i≤m (2)
each retransmission due to a collision, CW is doubled up to a
maximum value, Wm′ = CWmax = 2m′ ⋅ CWmin , where Wm' is the bi,0 = pi. b0,0 , 0<i≤m (3)
largest contention window size. Once the CW reaches CWmax, it Owing to the chain regularities and using (2), we have:
will remain at the value of CWmax until it is reset. Therefore, we
have: Wi − k
W
bi , k = ⋅ bi , 0 , i∈[0,m] , k∈[0,Wi-1] (4)
 i = 2 i ⋅W i ≤ m′ Wi

 W i = 2 m ′ ⋅W i > m′ (1)
Equations (3) and (4) express all bi,k values as a function of b0,0
where i is the backoff stage, i∈[0,m] and m represents the and p. If the normalization condition is imposed, finally b0,0 is
station short retry count. Here m is also the maximum backoff given by (5) and depends on the values of m and m'. Using the
stage. Our model considers that m can have a value larger or previous analysis, we can derive the probability τ that a station
smaller than m'. The 802.11b standard [4] specifies that m = 6 transmits a packet in a randomly chosen slot time. Since a
and m' = 5. station transmits when its backoff timer reaches the value of
After the successful reception of a packet in the destination zero, τ can be found as:
station, an immediate positive acknowledgment (ACK) is sent m m
1 − p m +1 (6)
back after a time interval equal to Short Inter-Frame Space τ = ∑ b i , 0 = ∑ p i ⋅ b0,0 = b0,0 ⋅
i =0 i =0 (1 − p )
(SIFS). Since SIFS is shorter than DIFS, the station sending an
ACK attempts transmission before stations attempting to send From (6) we can see that the transmission probability τ
new packets and hence takes priority. If the source station does depends on the collision probability p, which is derived next.

 2 ⋅ (1 − 2 p ) ⋅ (1 − p )
 W ⋅ (1 − ( 2 p ) m + 1 ) ⋅ (1 − p ) + (1 − 2 p ) ⋅ (1 − p m + 1 ) , m ≤ m′

b0,0 = (5)
 2 ⋅ (1 − 2 p ) ⋅ (1 − p )
 , m > m′
 W ⋅ (1 − ( 2 p ) m ′ + 1 ) ⋅ (1 − p ) + (1 − 2 p ) ⋅ (1 − p m + 1 ) + W ⋅ 2 m ′ ⋅ p m ′ + 1 ⋅ (1 − 2 p ) ⋅ (1 − p m − m ′ )

GLOBECOM 2003 - 951 - 0-7803-7974-8/03/$17.00 © 2003 IEEE

 The probability p that a transmitted packet encounters a A. Packet drop probability
collision is the probability that at least one of the n-1 The packet drop probability is defined, as the probability
remaining stations transmit in the same time slot. If all stations that a packet is dropped when the retry limit is reached and it
transmit with probability τ, the collision probability p is: is equal to:
p = 1 − (1 − τ ) n − 1 (7) pdrop = p m+1 (11)
Equations (6) and (7) form a nonlinear system with two since a packet is dropped if it encounters m+1 collisions.
unknowns τ and p. This nonlinear system can be solved
utilizing numerical methods and has a unique solution. Note B. Average time to drop a packet
that p∈[0,1] and τ∈[0,1].
A packet is dropped when it reaches the last backoff stage
Let Ptr be the probability that at least one station transmits
and experiences another collision. The average value that the
a packet in the considered slot time. For a wireless LAN of n
station will select for its backoff in the i stage, is (Wi+1)/2,
contending stations, Ptr is given by:
where i∈[0,m]. Let E[Tdrop] be the average number of slot
Ptr = 1 − (1 − τ )n (8) times required for a packet to experience m+1 collisions in the
(0,1,…m) stages:
The probability Ps that an occurring packet transmission is
W ⋅ (2m+1 −1) + (m +1)
successful is given by the probability that exactly one station  , m ≤ m′
m
W +1  2
transmits and the remaining n-1 stations defer transmission, E[Tdrop] = ∑ i =  '
m +1 m '
'
(12)
conditioned on the fact that at least one station transmits: i =0 2 W ⋅ (2 −1) +W ⋅ 2 ⋅ (m − m ) + (m+1) , m > m′
 2
n ⋅ τ ⋅ (1 − τ ) n −1 (9)
Ps = The average length of a slot time is:
1 − (1 − τ ) n
E[slot] = (1-Ptr)⋅ σ + Ptr .Ps .Ts + Ptr⋅(1-Ps) ⋅Tc (13)
Considering that a random slot is empty with
probability (1− Ptr ) , contains a successful transmission with Finally, the average time to drop a packet is equal to:
probability Ptr ⋅ PS and a collision with probability Ptr ⋅ (1− PS ) , E[Ddrop] = E[Tdrop] . E[slot] (14)
the saturation throughput S is given by:
C. Average packet delay
Ptr ⋅ PS ⋅ l (10) The average delay for a successfully transmitted packet is
S =
(1 − Ptr ) ⋅ σ + Ptr ⋅ PS ⋅ T S + Ptr ⋅ (1 − PS ) ⋅ T C defined to be the time interval from the time the packet is at
the head of its MAC queue ready to be transmitted, until an
where l is the length of the transmitted packet, σ the duration acknowledgement for this packet is received. If a packet is
of an empty slot time, Ts and Tc are the average times that the dropped because it has reached the specified retry limit, the
medium is sensed busy due a successful transmission or a delay time for this packet will not be included into the
collision respectively. The values of Ts and Tc depend on the calculation of the average delay.
channel access mechanism and in the case of basic access: The average packet delay E[D] , provided that this packet
is not discarded, is given by:
T Sb a s = D I F S + H + l + S I F S + A C K
T Cb a s = D I F S + H + l + S I F S + A C K E[ D] = E[ X ] ⋅ E[slot ] (15)
For the RTS/CTS access mechanism, it is: where E[X] is the average number of slot times required for
T SR T S = D I F S + R T S + S I F S + C T S + S I F S + H + l + S IF S + A C K successfully transmitting a packet and is given by:
T CR T S = D I F S + R T S + S I F S + C T S
 i m +1 W +1
where H = MAChdr + PHYhdr is the packet header.
m ( p − p )⋅ i
2  (16)
E[ X ] = ∑ 
1− p m +1 
i =0  
 
IV. PACKET DELAY ANALYSIS where (1- pm+1) is the probability that the packet is not dropped
Our analytical model considers the following metrics for and (pi - pm+1)/(1- pm+1) is the probability that a packet that is
the delay performance of IEEE 802.11 protocol, taking into not dropped reaches the i stage. After some algebra, (16)
account the retry limits of a data packet transmission. becomes equal to (17).

 W ⋅ (1 − (2 p ) m +1 ) ⋅ (1 − p ) + (1 − 2 p ) ⋅ (1 − p m +1 )
 -p m +1 ⋅ E[Tdrop ] , m ≤ m′
 2 ⋅ (1 − 2 p ) ⋅ (1 − p )
 (17)
E[ X ] = 
 m ' +1 m' m ' +1 '

 W ⋅ (1 − (2 p ) ) ⋅ (1 − p ) + W ⋅ 2 ⋅ p ⋅ (1 − p m − m ) ⋅ (1 − 2 p ) + (1 − 2 p ) ⋅ (1 − p m +1 )
-p m +1 ⋅ E [Tdrop ] , m > m′
 2 ⋅ (1 − 2 p ) ⋅ (1 − p )

GLOBECOM 2003 - 952 - 0-7803-7974-8/03/$17.00 © 2003 IEEE

 the average packet delay increases as the retry limit increases.
V. ANALYSIS RESULTS and that the packet delay decreases if a smaller retry limit than
Unless otherwise specified, the values reported in the the proposed value is employed, in both the basic access and
following figures have been obtained using the system RTS/CTS mechanisms. Especially in the case of medium or
parameters in table I and are based on the Direct Spread large network size (n > 15) the packet delay attains a lower
Sequence Spectrum (DSSS) physical layer used in 802.11b value but at the expense of more packets being dropped.
standard.
1.2
TABLE I DSSS SYSTEMS PARAMETERS IN 802.11b ◆ Basic (no retry limits) g RTS (no retry limits)

1 ‘ Basic, m=6 … RTS, m=6

Parameter Value S Basic, m=6 (simulation) U RTS, m=6 (simulation)

Packet delay (sec)

Packet payload, l 8184 bits 0.8
Slot time, σ 20 us
MAC header 224 bits 0.6

PHY header 192 bits

0.4
RTS packet 160bits + PHY header
ACK packet 112bits + PHY header 0.2
CTS packet 112bits + PHY header
DIFS 50 µs 0
5 10 15 20 25 30 35 40 45 50 55 60 65 70
SIFS 10 µs
Number of stations
Channel bit rate 1 Mbps
Minimum CW, W0 32 Fig. 2. Packet delay versus n for Basic access and RTS/CTS
Number of CW sizes, m' 5
Short retry limit 6
0.9 0.85
◆ Packet delay, m=6 ‘ Packet delay, m=4
0.8 0.8
Fig. 2 plots the packet delay against the number of stations S Throughput, m=6 U Throughput, m=4
and validates our analytical model since an almost exact match 0.7 0.75

Throughput efficiency
between analytical and simulation results is observed.
Packet Delay (sec)

0.6 0.7
Moreover, this figure reports the different packet delay values 0.5 0.65
obtained from our model compared to results from a model 0.4
Throughput
0.6
that does not use any retry limits [12]. In both the cases of
0.3 0.55
basic access and RTS/CTS mechanisms, the packet delay
attained is higher when there are no retry limits for a packet 0.2
Delay
0.5

transmission. This is easily justified by noting that in the case 0.1 0.45
of no retry limits, a packet will be transmitted continuously 0 0.4
until its successful reception. 5 10 15 20 25 30 35 40 45 50 55 60 65 70
The dependency of the average packet delay on the Number of stations

number of stations and the retry limit is examined in both the Fig. 3. Throughput efficiency and packet delay for basic access
cases of the basic access and RTS/CTS mechanisms
respectively in fig. 3 and 4. Note that in fig.3, the vertical axis
for the throughput efficiency is different to that in fig. 4. Both 0.6 0.85
figures show that the average packet delay is an increasing ◆ Packet delay, m=6 ‘ Packet delay, m=4
0.84
function of the number of stations. This occurs because (i) a 0.5 S Throughput, m=6 U Throughput, m=4
station has to wait longer due to the successful transmissions
Throughput efficiency

0.83
Packet Delay (sec)

0.4
of other stations and (ii) more contending stations increase the
probability of collisions and the number of retransmissions. 0.82
0.3
The RTS/CTS mechanism provides a lower delay value than Throughput
0.81
the basic access due to the smaller length of the colliding 0.2
packets. 0.8

Also, fig. 3 and 4 report throughput and packet delay 0.1 Delay
0.79
values for two different packet retry limits and for both basic
access and RTS/CTS mechanisms. Note that the IEEE 802.11 0 0.78
standard proposes the value 6 for the retry limit. Results show 5 10 15 20 25 30 35 40 45 50 55 60 65 70
Number of stations
that the retry limit considerably affects the throughput
performance of the 802.11 protocol. Both figures illustrate that Fig. 4. Throughput efficiency and packet delay for RTS/CTS

GLOBECOM 2003 - 953 - 0-7803-7974-8/03/$17.00 © 2003 IEEE

 9 0,16 model, which takes into account packet retry limits, is
8 0,14
compared to a model without any retry limits. Comparison
7
with simulation results showed that our model can predict

Packet Drop probability

0,12
accurately the delay performance metrics of the 802.11
Packet Drop time

6
0,1 protocol. We also have shown that using packet retry limits
)

5
0,08
results in a lower average delay for a packet transmission
comparing to the case without retry limits, however, at the
(

4
Drop time
3
0,06 expense of finite packet drop probability.
0,04
2

1 0,02 REFERENCES
0 0 [1] IEEE standard for Wireless LAN Medium Access Control (MAC)
5 10 15 20 25 30 35 40 45 50 55 60 65 70 and Physical Layer (PHY) specifications, ISO/IEC 8802-11:1999(E),
Number of stations Aug. 1999.
[2] HiperLAN- High Performance Radio Local Area Network Draft ver.
Drop time Drop probability 1.1, ETSI, 1995.
◆ Basic, m=6 ‘ RTS, m=6 Both Basic and RTS [3] Wireless LAN Medium Access Control (MAC) and Physical Layer
S Basic, m=4 U RTS, m=4 g m=4 … m=6 (PHY) Specification: High-Speed Physical Layer in the 5 GHz Band,
IEEE 802.11a WG, Sept. 1999.
Fig. 5. Drop time and Drop probability versus n [4] Wireless LAN Medium Access Control (MAC) and Physical Layer
(PHY) Specification: High-Speed Physical Layer Extension in the
2.4 GHz Band, IEEE 802.11b WG, Sept. 1999.
Fig. 5 depicts the effects of the retry limit on the packet [5] [Link], [Link]. “Interference Analysis of Nonpersistent CSMA
drop probability and the packet drop time. Some with Hidden Terminals in Multicell Wireless Data Networks”, Proc.
considerations are presented regarding the average drop time, PIMRC, Toronto, pp.907-911, 1995.
[6] H.S. Chhaya, S. Gupta, “Performance modeling of asynchronous
namely the time in which a packet will be dropped because it data transfer methods of IEEE 802.11 MAC protocol”, Wireless
reaches the maximum the retry limit. In all cases, the Networks, pp.217-234, 1997.
RTS/CTS mechanism achieves a lower value for the average [7] Y. C. Tay , K. C. Chua, ”A capacity analysis for the IEEE 802.11
drop time, with respect to the basic access mechanism. This MAC protocol”, Wireless Networks, Volume 7, Issue 2 , pp. 159-
171, March 2001.
decrease is mainly observable when the network size n leads [8] J. Weinmiller, H. Woesner, JP Ebert, A. Wolisz, "Analyzing and
to a higher collision probability (n = 70). Furthermore, fig. 5 Tuning the Distributed Coordination Function in the IEEE 802.11
allows us to answer the question on the dependence of the DFWMAC Draft Standard", Proc. of MASCOT‘96, San Jose,
average drop time on the retry limit. A small value of the retry California, 1996.
[9] Marek Natkaniec, Andrzej R. Pach, “Analysis of the Backoff
limit (m = 4), results in a low average drop time. A higher mechanism in IEEE 802.11 standard”.,IEEE Symposium on
value for the retry limit, (m = 6), corresponds to the higher Computers and Communications, Antibes, France, July 2000.
average drop time of 8.4 s for n = 70 if the basic access [10] [Link], [Link], [Link]. “Dynamic Tuning of the IEEE 802.11
method is used. Protocol to Achieve a Theoretical Throughput Limit”, IEEE/ACM
Trans. On Networking, V8, N6, 2000.
Moreover, as shown in (11), the value of the packet drop [11] H. Wu,, K. Long, S. Cheng, J. Ma, “IEEE 802.11 Distributed
probability depends on the retry limit and the collision Coordination Function (DCF): Analysis and Enhancement”,
probability. Since the packet drop probability does not depend International Conference on Communications (ICC), April 2002.
on access mechanism, the results presented in fig.4 are [12] [Link]. “Performance Analysis of the IEEE 802.11 Distributed
Coordination Function”, IEEE Journal on Selected Area in Comm.
applicable on both basic access and RTS/CTS. V18, N3, pp. 535-547, 2000.
More specifically, when the retry limit is equal to the [13] H. Wu, Y. Peng, K. Long, S. Cheng, J. Ma, “Performance of Reliable
proposed value of the standard (m = 6), the packet drop Transport Protocol over IEEE 802.11 Wireless LAN: Analysis And
probability increases as the number of stations increases. Enhancement”, IEEE INFOCOM'2002.
[14] [Link], [Link], A. C. Boucouvalas, “Throughput and
However, for small values of the retry limit and a large Delay analysis of IEEE 802.11 protocol” Proc. of IEEE International
network size, the packet drop probability increases rapidly. As Workshop on Networked Appliances (IWNA), Liverpool, UK, pp.
a result, a packet drop probability of 0.14 is obtained if the 168-174, 2002.
retry limit is m = 4 for a network of 70 contending stations.

VI. CONCLUSIONS
In this paper, an analytical model using a Markov chain
was developed to compute the packet delay, the packet drop
probability and the packet drop time protocol of IEEE 802.11
protocol. This analysis can be used in both the basic access
and RTS/CTS methods and includes retry limits. Our proposed

GLOBECOM 2003 - 954 - 0-7803-7974-8/03/$17.00 © 2003 IEEE