2017 14th International Symposium on Pervasive Systems, Algorithms and Networks & 2017 11th International Conference
on Frontier of Computer Science and Technology & 2017 Third International Symposium of Creative Computing
Performance Analysis of WLANs under Bursty and
Correlated Video Traffic
Noushin Najjari, Geyong Min, Jia Hu, Wang Miao
College of Engineering, Mathematics and Physical Sciences, University of Exeter, UK
Email: {nn248, g.min, j.hu, wm255}@exeter.ac.uk
Abstract- Proliferation of mobile digital devices has posed
major challenges for design and implementation of Wireless
Local Area Networks (WLANs), leading to stringent quality
of service (QoS) requirements. Extreme burstiness and
correlation characteristics of multimedia traffics generated
by digital devices require accurate modelling and analysis to
provide satisfactory end user experience. Most existing
performance models for WLANs are developed under
unrealistic assumptions where burstiness and correlation
characteristics of network traffic are ignored. This paper
proposes an original analytical model as a cost-effective tool
to evaluate the performance of WLANs in the presence of
realistic video traffic, taking into account burstiness and
correlation characteristics using Batch Markovian Arrival
Process (BMAP). The model derives important QoS metrics
in terms of throughput, end-to-end delay and frame loss
probability. The accuracy of the obtained results are
validated via extensive simulation experiments executed in
NS2 subject to traffic parameters obtained from real-world
video sources generated by practical multimedia
applications. The results reveal how burstiness and
correlation of video traffic affects QoS performance of
WLANs.
Keywords- Analytical modelling, Bursty and Correlated
Traffic, Batch Markovian Arrival Process, IEEE 802.11
WLAN, Quality-of-Service.
I. INTRODUCTION
Rapid development of wireless communications and
networking technologies along with the explosive growth
in the number of mobile devices such as smart phones and
personal digital assistants have generated extensive
attention on the study of wireless multimedia traffic and
it’s transmission. Several technology forecasts predict that
by 2020, 75% of the world’s mobile data traffic will be
video [1]. The great strain of video traffic will be a stern
test for the 802.11-based Wireless Local Area Networks
(WLAN) which have gained widespread popularity due to
their low cost and easy deployment.
Video traffic has much higher bit rates than other types
of traffic transmitted through WLANs. Many high quality
measurement studies [2, 3] have shown that realistic traffic
generated by multimedia applications, and in particular
traffic generated by means of video applications in
wireless and mobile networks, exhibits self-similar nature,
bursty and correlated outlook over a wide range of
timescales. The use of miscellaneous encoding schemes by
different applications for content delivery assurance of
video traffic as well as energy consumption reduction has
resulted in such traffic to develop specific characteristics
of burstiness, self-similarity and correlation [4, 5]. The
coding mechanisms use Rate Adaptive techniques in order
to adapt the transmission rate and packet sizes of video
traffic with the capacity of the wireless medium. As a
result, video data is broken into different sized packets and
transmitted with different rates to provide higher Quality-
2375-527X/17 $31.00 © 2017 IEEE
DOI 10.1109/ISPAN-FCST-ISCC.2017.52
of-Service (QoS) for the end users. Hence, accurate
modelling of traffic needs not only capture packet arrival
rates but should also closely match variable packet size
distributions. This will in turn help towards accurate
analysis of degrading factors in WLANs, such as long
delays, severe packet dropping and large buffer
requirements which are the outcome of burstiness,
correlation and self-similarity.
Many studies still rely on conventional teletraffic
models, such as Poisson Process, for modelling the traffic
within networks. However these models, although initially
successful and analytically simple, are no longer accurate
and adequate for capturing traffic characteristics of today’s
applications which are much richer in content and more
diverse in traffic patterns. Batch arrivals, event correlations
and burstiness of traffic generated by compressed video in
modern communication networks are of paramount
importance as latency or delay plays a critical role in user
experience [2, 6, 7]. To develop a more accurate analysis
and in an effort to bridge the existing gap in literature for
capturing burstiness and correlation properties of modern
multimedia traffic, this paper proposes a novel analytical
model for evaluating the performance of IEEE 802.11-
based WLANs under video traffic using the Batch
Markovian Arrival Process (BMAP) [8]. BMAP has the
ability to capture different packet sizes and traffic rates
while accurately modelling aggregated traffic which often
possesses negative statistical properties [9]. Major
contributions of this paper are as follows:
• A versatile and comprehensive analytical model is
proposed for modelling IEEE 802.11-based
WLANs under bursty and correlated traffic. The
model is developed using BMAP and real network
conditions, e.g. unsaturated network stations, retry
limit and limited buffers.
• To capture and compare the traffic characteristics
of stations in different WLANs, Poisson Process is
employed to model the traffic of non-bursty data
applications, Markov Modulated Poisson Process
(MMPP) is used to model the traffic of ON/OFF
voice applications and BMAP is used to model the
bursty traffic of video applications.
• QoS performance metrics in terms of throughput,
frame loss probability and end-to-end delay are
derived and validated using extensive simulation
experiments carried out in NS2. Simulation results
reveal great accuracy and similarity with those
obtained from the analytical model.
The remainder of this paper is structured as follows.
Section II provides an overview of existing literature on
analytical modelling of 802.11 WLANs under multimedia
traffic. Section III presents an overview of the 802.11-
based WLAN, as well as the important properties of
BMAP. This is followed by the presentation of the new
analytical model in section IV. Section V presents in
255
250
details the simulation experiment setup, accuracy
validation of the model and performance evaluation.
Finally, section VI concludes the paper.
II. RELATED WORK
Modelling the performance of wireless networks under
realistic traffic has been a topic of interest in research for a
long time. Today’s WLANs are integrating and
transmitting a diverse range of traffic generated by
multimedia applications which significantly differ with
each other in packet arrival rates and packet sizes,
including video, voice and text [1, 10]. Many of the
existing works in literature have focused on the use of
Poisson Process to model the traffic of WLANs in order to
obtain various performance measures irrespective of
realistic network traffic characteristics [11, 12]. For
example, in [13], M/G/1 queue is used to model and
analyse 802.11-based WLANs under network saturation
condition. In [14], the authors describe the behaviour of
each station in 802.11-based WLANs as a Markov renewal
process by modelling each station as a Ȁ
Ȁͳ queue. The retransmits the frame. Each station maintains a retry
main interest for the use of Poisson process is its
tractability and simplicity. Finding a process that can
accurately capture the specific characteristics of traffic
within networks, while remaining analytically tractable is
not easy. With multimedia traffic and in particular video
forming most of the traffic in WLANs, many researchers
have focused on the use of such processes as the Markov
Modulated Poisson Process (MMPP) [15] to capture traffic
burstiness and correlation. MMPP is used extensively in
literature to accurately model the voice traffic [11, 16, 17].
In [11] the authors used the superposition of a number of
two-state MMPPs to model the ON-OFF characteristics of
voice traffic in wireless multimedia networks. However,
MMPP is only able to capture the burstiness and
correlation of multimedia traffic, and lacks the ability to
capture the inherent self-similarity characteristic that exists
in video traffic. Besides self-similarity, video traffic
requires models that can capture the existing correlation
between packet size distribution as well as packet arrival
rates. Therefore, MMPP proves to be inadequate as it is
only able to capture the correlation and burstiness
characteristics through modelling of variable arrival rates.
BMAP on the other hand, can model dependent and non-
exponential inter-arrival time distributions between
batches and correlated batch sizes of arriving packets with
varying arrival rates. Despite some studies concentrating
on the analysis of WLANs under multimedia traffic [12,
18, 19]; to the best of our knowledge, none have focused
on the use of BMAP for the performance evaluation of
ሺ–ሻሽ on the state space of ሼሺ‹ǡ Œሻǣ ‹ ൒ Ͳǡͳ ൑ Œ ൑
802.11-based WLANs under multimedia traffic.
III. PRELIMINARIES
A. Wireless Local Area Networks
As a result of massive changes in consumption patterns
of digital devices, large amounts of traffic are constantly
generated and transferred on ubiquitous IEEE 802.11-
based WLANs. The main channel access mechanism
provided by 802.11 is the Distributed Coordination
Function (DCF), which allows sharing of the wireless
medium through the use of Carrier Sense Multiple Access
with Collision Avoidance (CSMA/CA). In CSMA/CA, a
station senses the channel before attempting to transmit
251
256
any frames. Transmission starts only if the channel is
sensed idle for the period of time of Distributed Inter-
frame Space (DIFS). Otherwise, the node persists
monitoring the channel activity until sensing an idle
channel for a DIFS period and then generates a random
backoff interval before transmission. The backoff time is
uniformly chosen from the range ሾͲǡ ୧ିଵ ሿ, where ୧ is
the current contention window size and ‹ is the backoff
stage. After each unsuccessful transmission, ୧ is doubled,
ᇲ
up to a maximum value of ୫ᇲ ൌ ʹ୫ , where ᇱ
represents the maximum backoff stage. The backoff
counter is decremented by one for each time slot when the
channel is sensed idle, “frozen” when the channel becomes
busy and reactivated when the channel is sensed idle again
for a DIFS. A station transmits a frame when its backoff
counter reaches zero. Upon the successful reception of the
frame, the destination station sends back an
Acknowledgement (ACK) frame immediately following a
Short Inter-frame Space (SIFS) interval. If the station does
not receive the ACK within a timeout interval [20], it
counter that is increased by one after each retransmission.
The frame is discarded if the number of retransmission
attempts reaches the retry limit, . The DCF mechanism is
part of the Medium Access Control protocol (MAC) in the
TCP/IP protocol stack. MAC protocol is one of the most
significant aspects of any ad-hoc network as it has a direct
bearing on the efficiency and reliability of data
transmission.
B. Batch Markovian Arrival Process:
Bursty and correlated traffic generated by multimedia
applications can considerably deteriorate the user
perceived QoS and have considerable effect on the
exhibited performance of WLANs for different 802.11
variants [21]. To capture the correlation and burstiness, we
propose the use of Batch Markovian Arrival Process
(BMAP) [8, 22]. BMAP is a subclass of Stochastic Point
process and generalizes such known flows as the stationary
Poisson Process, Phase type process (PH), Markov
Modulated Poisson Process (MMPP), Interrupted Poisson
Process (IPP), Markovian Arrival Process (MAP), etc, and
has been extensively studied and applied to various real
world communication systems [23, 24]. BMAP generalizes
the standard Point Processes by allowing “batches” of
arrivals, dependent inter-arrival times, non-exponential
inter-arrival time distributions between batches and
correlated batch sizes. The behaviour of BMAP [8] is
defined by an underlying two dimensional Markov Process
of ሼሺ–ሻǡ
ሽ. The  ൈ  matrices represented as ୩ ǡ  ൒ Ͳ, define the
transitions of the Markov Chain accompanied by arrival of
batches consisting of  packets. Matrix ଴ has negative
diagonal elements and non-negative off-diagonal elements,
where the absolute value of the negative diagonal elements
represents the arrival rates of respective states of the
Markov Process. For  ൒ ͳ and ୩ ൒ Ͳ, the irreducible
infinitesimal generator  of the process is defined as:
ሺœሻ ൌ σஶ ୩
୩ୀ଴ ୩ œ . Vector Ɏ is the steady state probability
distribution of the Markov process with generator ;
which satisfies:
ߨ‫ ܦ‬ൌ Ͳܽ݊݀ߨ݁ ൌ ͳ (1)
where ‡ is a column vector of ͳ’s and Ͳ is a row vector
consisting of all zeroes. The average (fundamental) arrival
rate of BMAP is calculated as follows:
ߣ௧௢௧ ൌ ߨ σஶ
௞ୀଵ ݇‫ܦ‬௞ ݁ ൌ ߨ݀ ǡ݀ ൌ σ ݇‫ܦ‬௞ ݁ (2)
C. Parameter Estimation for BMAP
BMAP has enough flexibility to describe a wide
variety of data, and rate fluctuations in many applications.
However a challenging issue in the study and application
of BMAP is the accurate estimation of its parameters based
on real-world applications. Any method used to estimate
the parameters has to be as accurate as possible and should
keep the number of states small enough for a tractable
model. Due to incomplete data, standard statistical
techniques such as moment matching cannot be used for
BMAP. As measured trace data does not contain all
statistical properties required for the unique specification
of a corresponding BMAP, a very common and important
method for its parameter estimation would be the
Maximum-Likelihood Estimation method (MLE). In this
regard; the Expectation-Maximization (EM) algorithm is
one of the methods of computing MLEs effectively. It is a
statistical framework that computes the MLEs under
incomplete data and is particularly useful for stochastic
models with many parameters, hence BMAP. Limited
number of works exists in literature which focuses on
developing BMAP parameters from existing data traces
[25, 26]. This paper uses the technique developed in [25]
which is based on EM algorithm, for parameter estimation
of BMAP from actual data traces of video traffic.
IV. THE ANALYTICAL MODEL
In this section the analytical model for evaluating the
performance of unsaturated WLANs under video traffic
using the Batch Markovian Arrival Process is presented.
The transmission queue at each station is modelled as a
Ȁ
ȀͳȀ queuing system where the arrival traffic
follows a BMAP with the average arrival rate of ɉ୲୭୲
(frames/second), to support video traffic.
A. Analysis of the Service Time
Bianchi [27] developed a bi-dimensional Markov chain
to model the backoff procedure of the IEEE 802.11 in
single hop WLANs, deriving the saturation transmission
probability, with the assumption that all stations are always
ready for transmission and their transmission queues are
assumed to always be non-empty. D. Malone et al. in [28],
extend Bianchi’s model for non-saturated conditions and
Wu et al in [29] extend the model to accommodate the case
of retry limit. In this section, the models of [28, 29] are
extended to develop an analytical model for WLAN under
unsaturated network conditions with limited retry and
limited buffer size in order to develop a condition closer to
realistic networks for applying the video traffic. We
assume that the WLAN is composed of fixed number of
identical stations ሺሻ. A station transmits only when its
transmission queue is non-empty, therefore the
transmission probability ɒ is calculated by weighting the
saturation transmission probability with the probability of
the non-empty transmission queue:
߬ ൌ ሺͳ െ ܲ଴ ሻ߬ ᇱ (3)
where ଴ is the probability that the transmission queue of
the station is empty. Eq.(3) is calculated with the
assumption of no post backoff as stated in [28]. ɒᇱ is the
saturation transmission probability given as [29]:
߬ᇱ ൌ
ଶሺଵିଶ௣ሻሺଵି௣ሻ
ᇲ ᇲ ᇲ ᇲ
ሺଵିଶ௣ሻሺଵି௣೘శభ ሻାሺଵି௣ሻௐ൫ଵିሺଶ௣ሻ೘ శభ ൯ାௐଶ೘ ௣೘ శభ ሺଵିଶ௣ሻሺଵି௣ሻ೘ష೘
(4)
In Eq.(4),  is the minimum contention window size,
 is the maximum backoff stage (i.e. retry limit) and
ᇱ denotes the maximum number of times that  can be
doubled. ’ is the conditional collision probability and is
equal to the probability that at least one of the remaining
stations transmits in a given time slot:
‫ ݌‬ൌ ͳ െ ሺͳ െ ߬ሻ௡ିଵ (5)
The mean service time is the summation of the average
channel access delay, ሾሿǡand average transmission
delay,ୱ . Thus, the mean service time is calculated as:
‫ܧ‬ሾܵሿ ൌ ‫ܧ‬ሾ‫ܣ‬ሿ ൅ ܶ௦ (6)
Assuming that the frame is successfully transmitted
after experiencing ŒሺŒ ൒ Ͳሻ collisions, its channel access
delay would equal to the delay caused by Πunsuccessful
transmissions and the ሺŒ ൅ ͳሻ back off stages, and can be
calculated as:
‫ܧ‬ሾ‫ܣ‬ሿ ൌ ܶ௖ ߮ ൅ ߪ ᇱ ߜ (7)
where ɐᇱ is the average size of a time slot and is calculated
differently at different stages. ୡ is the collision time, ɔ is
the average number of collisions before a successful
transmission from the station, ɐ is the average length of a
time slot, and Ɂ represents the average number of time
slots the station defers during backoff stages:
௝௣ೕሺଵି௣ሻ ௝ ௐ೓ ିଵ ௣ೕሺଵି௣ሻ
߮ ൌ σ௠ ௠
௝ୀ଴ ሺଵି௣೘శభ ሻ ߜ ൌ σ௝ୀ଴ σ௛ୀ଴ ሺଵି௣೘శభ ሻ
(8)
ଶ
In Eq.(8), ’୨ is the probability that the frame
୛ ିଵ
experiences ŒሺͲ ൑ Œ ൑ ሻ collisions, and ሺ ౞ ሻ denotes
ଶ
the mean of the backoff counters generated in the Š-th
ሺͲ ൑ Š ൑ Œሻ backoff stage. ୱ is the probability that a
transmission occurring on the channel is successful given
by the probability that exactly one station transmits on the
channel, conditioned on the fact that other stations are in a
backoff procedure:
ܲ௦ ൌ ݊߬ሺͳ െ ߬ሻ௡ିଵ ሺ9)
When the station is in backoff, the size of the time slot
is obtained by considering the fact that the channel is idle
with probability:
ܲ௜ௗ௟௘ ൌ ሺͳ െ ‫݌‬ሻ   ሺͳͲሻ
When the transmission is successful, it equals to ୱ .
And finally when there is a collision, the size of time slot
would equal to ሺ’ െ  ୱ ሻ. Therefore the value of ɐᇱ is
252
257
calculated as follows, where ɐ is the duration of an empty
physical time slot as mentioned in [20]:
ߪ ᇱ ൌ ሺͳ െ ‫݌‬ሻߪ ൅ ܲ௦ ܶ௦ ൅ ሺ‫ ݌‬െ ܲ௦ ሻܶ௖ 
, of the underlying CTMC of Figure (2) can
where ୱ is the average time the channel is sensed busy by
each station because of successful transmission, and ୡ is
the average time the channel is sensed busy by each station
during a collision [27]:
and the following equations can be used:
arrivals of batch size of , then the transmission rate would
be determined by ୩ ሺɄǡ Ʉ ൅ ͳሻ. The transition rate out of
state Ʉ, e.g. from ሺɄǡ •ሻ to ሺɄǡ • െ ͳሻ, would equal to the
(11) service rate Ɋ. Following the above analysis, the transition
rate matrix,
be obtained. To calculate the steady state probability
vector  ൌ ൫ୱǡ஗ ǡ • ൌ Ͳǡͳǡ Ǥ Ǥ Ǥ ǡ ǡ Ʉ ൌ Ͳǡͳǡʹ൯ ൌ
ሺ଴ ǡ ଵ ǡ ǥ ǡ ୒ ሻ of the Markov chain, the transition rate
matrix

ܶ௦ ൌ ܶு ൅ ܶ௅ ൅ ܵ‫ ܵܨܫ‬൅ ʹ߂ ൅ ‫ ܭܥܣ‬൅ ‫ܵܨܫܦ‬ ܲ‫ ܩ‬ൌ Ͳܽ݊݀ܲ݁ ൌ ͳ (13)

ܶ௖ ൌ ܶு ൅ ܶ௅ ൅ ‫ ܵܨܫܦ‬൅ ߂ (12) whereୱ ൌ  ൫ୱǡ଴ ǡ ୱǡଵ ǡ ୱǡଶ ൯ǡͲ ൑ • ൑ , and ‡ is a
column vector of ͳ’s.
ୌ is the average time required to transmit the packet
header. ȟ is the propagation delay and ୐ is the average C. Performance Measures
time required to transmit the longest packet payload. Ɏୱ denotes the steady-state probability of • frames
being in the queuing system of a station and is given by:
B. BMAP Queueing Analysis of Stations
As mentioned earlier, in this paper the stations of the ߨ௦ ൌ ܲ௦ ݂݁‫ Ͳݎ݋‬൑ ‫ ݏ‬൑ ܰ (14)
WLAN are modelled as ȀȀͳȀ queuing systems
[22], where  represents the limited buffer size of each The probability that the transmission queue of the
station. The idea is the same as any queueing model, when station is empty, ଴ , can be calculated by Ɏ଴ . Assume Ɏᇱୱ
a frame reaches the head of the transmission queue, the represents the steady-state probability of • frames being in
server becomes busy, and as soon as a frame is the queuing system of a station when a frame arrives. In
acknowledged by the destination following a successful the case of BMAP traffic, Ɏᇱୱ can be written as:
transmission, the server becomes free. The service time is
dependent on the size of the frame transmitted and is ௉ ௸௘
modelled by an exponential distribution function with ߨ௦ᇱ ൌ σಿ ೞ ǡ݂‫ Ͳݎ݋‬൑ ‫ ݏ‬൑ ܰ   ሺͳͷሻ
ೞసబ ௉ೞ ௸௘
mean ሾሿ. Thus, the service rate, Ɋ, can be calculated as:
Ɋ ൌ ͳȀሾ•ሿ.
BMAP is characterized by an underlying Continuous
Time Markov Chain (CTMC). The number of states in the
CTMC represents the number of states considered for
BMAP. At this stage, as regards to the practical size limits
and for ease of modelling and simulation, the BMAP of the
queueing model at each station of the WLAN is considered
to be composed of three states. The stations are assumed to
be able to generate batches of maximum three and five
frames at one time for different scenarios; however the
accuracy of the model will slightly increase as the CTMC
is extended, and a three state CTMC at this stage will
generate a good result. Figure (1) presents the ͵-state Figure 1 Three-state CTMC underlying BMAP with batch size three
CTMC of the model with maximum batch size of ͵, and
Figure (2) shows the state transition diagram of the
ȀȀͳȀ queue assuming that the maximum buffer
size of each station is ሺ ൌ ͷͲሻ and again the maximum
batch size is ͵. State ሺɄǡ •ሻ, where ሺ• ൌ Ͳǡ ǥ ǡ ሻ and
ሺɄ ൌ Ͳǡͳǡʹሻ, represent the case that there are • frames in
the queuing system and the three-state BMAP
characterizing the traffic of the station is at state Ʉ. The
BMAP is characterized based on the states and the batch
size of the arrival process. If the batch size is one, meaning
that only one frame arrives at the station at one time, then,
the transmission rate from state ሺɄǡ •ሻ to ሺɄǡ • ൅ ͳሻ in the
CTMC would be determined by the respective value from
the matrix of ଵ ሺɄǡ Ʉሻ, considering that the arrival rate of
the generated frame stays the same (the CTMC is still at
the same state). For batch sizes greater than ͳ, the
transmission rate from state ሺɄǡ •ሻ to ሺɄǡ • ൅ ሻ, would be
Figure 2. State transition diagram of the BMAP/M/1/N (N=50) queue
determined by ୩ ሺɄǡ Ʉሻ for ሺ ൐ ͳሻ, again with the arrival
rate staying the same as before. However if there is a where Ȧ is a diagonal matrix containing the arrival
transition from state Ʉ to any other state, e.g. Ʉ ൅ ͳ, with rates of each state of the underlying CTMC of BMAP. The

253
258
 values of the Ȧ matrix can be gained from the absolute
values of the diagonal elements of the ଴ matrix of
BMAP. The loss probability, is shown by ୠ , and can be
gained from Ɏᇱ୒ as:
௉ ௸௘
ܲ௕ ൌ ߨேᇱ ൌ σಿ ಿ ǡ݂‫ Ͳݎ݋‬൑ ‫ ݏ‬൑ ܰ
ೞసబ ௉ೞ ௸௘
With ሾሿ being the frame payload size and ɉ୲୭୲ the
fundamental traffic arrival rate of the station, the
throughput,, of the station can be computed by:
ܶ‫ ܪ‬ൌ ߣ௧௢௧ ‫ܧ‬ሾܲሿሺͳ െ ܲ௕ ሻ   ሺͳ͹ሻ
Using the Little’s Law, the average end-to-end delay of
a frame being transmitted from a source to a destination,
ሾሿ, can be calculated as:
ாሾேሿ
‫ܧ‬ሾ‫ܦ‬ሿ ൌ
ఒ೟೚೟ ሺଵି௉್ ሻ
The end-to-end delay is the time interval from the
instant that a frame enters the transmission queue of the
source station, to the instant that the frame is
acknowledged after successful transmission by the
destination station. ሾሿ is the average number of frames
in the queuing system of the station, which can be
computed as:
‫ܧ‬ሾܰሿ ൌ σே
௦ୀଵ ‫ߨݏ‬௦   
V. MODEL VALIDATION AND PERFORMANCE EVALUATION
The accuracy of the developed analytical model has
been validated through extensive simulations using the
NS2 simulation environment. For this purpose a new
traffic generator is developed in NS2 to accommodate the
Batch Markovian Arrival Process using C++ and TCL
programing. The practical usefulness of the analytical
results using BMAP depends on how good the model is
parameterised based on the original traffic trace using the
EM algorithm presented in [25]. To generate correlations
between variable packet sizes and arrival rates, at this stage
a three state CTMC for the BMAP with maximum batch
sizes of three and five are assumed. For flexibility and
scalability of the simulation experiments and the analytical
model, real-world video trace obtained from high quality
video stream measurements of the film “Tears of Steel”,
encoded in H.265/HEVC codec [30] is used. Table I
presents the some information on the data trace.
Also using the same technique, the parameters for a
three-state MMPP are estimated. To vary the traffic load of
the stations, the arrival rates of the ଴ matrix are varied
according to the overall load of the traffic placed on the
network. The infinitesimal generator matrix of the three-
state BMAP with batch size of maximum three, , and
matrix ଴ gained are as follows:
TABLE I. DATA TRACE INFORMATION
Sequence Tears of Steel
Resolution ͳͻʹͲ ൈ ͺͳ͸
Frames Per Second ʹͶ
Number of Frames ͳ͹ͷͻʹ
Average Frame Size (Bytes): ͻͺͺǤͷͲ
െߣଵ ͲǤͲͳ ͲǤͲ͵ െͲǤͳ͹ ͲǤͲ͹ ͲǤͳͲ
‫ܦ‬଴ ൌ ͲǤͲͷ െߣଶ ͲǤͲͺܳ ൌ ͲǤ͵Ͷ െͲǤ͹ ͲǤ͵͸
ͲǤͳ͹ ͲǤʹͲ െߣଷ ͲǤ͸ͳ ͲǤ͹ െͳǤ͵͵
where ɉଶ ൌ ͲǤͶ ‫ כ‬ɉଵ and ɉଷ ൌ ͲǤͳ ‫ כ‬ɉଵ and ɉଵ ൌ ͲǤͷ ‫ כ‬ɉ,
(16) with ɉ being the overall traffic load. For three-state BMAP
with maximum batch size of five the following results
were estimated using the same data trace:
െߣଵ ͲǤͲ͵ ͲǤͲ͵ െͲǤʹͷ ͲǤͳͶ ͲǤͳͳ
‫ܦ‬଴ ൌ ͲǤͳ െߣଶ ͲǤͳ ܳ ൌ ͲǤͶͻ െͳǤͳ ͲǤͷ͹
ͲǤʹ ͲǤͳʹ െߣଷ ͲǤ͹ͺ ͲǤͻ͸ െͳǤͻ͵
For the purpose of comparison, the developed model is
executed using BMAP, MMPP and Poisson processes. To
generate comparable results, the overall load of the stations
is kept the same in all the models. The stationary
probability vector of all three models using BMAP and
(18) MMPP is Ɏ ൌ ሾͲǤ͹ʹʹ ͲǤͳ͹ͷ ͲǤͳͲʹሿ. With the
infinitesimal generator  and ଴ matrices of the BMAPs
defined as above, the rest of the matrices, ଵ to ୩ ሺ ൌ
͵‘”ͷሻ , are developed using the method presented in [23].
For three-state MMPP the following was estimated for the
infinitesimal generator  using the EM algorithm:
െͲǤͲ͹ͻ ͲǤͲ͵ͻ ͲǤͲͶ
ܳ ൌ ͲǤͳͺͲ െͲǤͶͳ͸ ͲǤʹͶ
ͲǤ͵͵ ͲǤ͵͵ െͲǤ͸͸
(19) The MMPP with infinitesimal generator  and arrival
rate matrix Ȧ ൌ †‹ƒ‰ሺɉଵǡǥǡ ɉ୫ ሻ is a BMAP process [22],
where ଴ ൌ  െ Ȧ, and ଵ ൌ Ȧ , and ୩ ൌ Ͳሺ ൐ ͳሻ.
Therefore with the infinitesimal generator estimated above,
and the arrival rates set as ɉଶ ൌ ͲǤͶ ‫ כ‬ɉଵ , ɉଷ ൌ ͲǤͳ ‫ כ‬ɉଵ and
ɉଵ ൌ ͲǤͷ ‫ כ‬ɉ, the result of the models will be comparable.
The corresponding Poisson process in the format of BMAP
is a process with rate ɉ, where ଴ ൌ െɉ and ଵ ൌ ɉ.
A. Simulation scenario
To validate the results, the developed model is
simulated within a Basic Service Set (BSS) of WLANs
where ሺ ൌ ͳͲሻ static stations are distributed within a
rectangular ͳͲͲͳͲͲ grid, and each station generates
and transmits traffic through a single-hop route to its
paired stations. Video traffic is then injected into the MAC
buffer of the stations.
In the simulation scenario, the stations are equipped
with the 802.11b physical layer and ideal wireless channel
conditions are assumed. The buffer size of all stations is
configured to be maximum 50 frames and the size of all
data frame payload is set to be 500 Bytes. The remaining
simulation settings of the stations and the WLAN are
summarized in Table II.
Each simulation is executed for 600 seconds of NS2
simulation time, which is sufficiently long to gain a stable
simulation and reliable performance results. To be able to
show the accuracy of BMAP in modelling network traffic,
four scenarios are considered: i) traffic generated by the
stations modelled as a 3-state BMAP with maximum batch
size of 3, ii) traffic generated by the stations modelled as a
3- state BMAP with maximum batch size of 5, iii) traffic
generated by the stations modelled using a 3-state MMPP,
and iv) traffic generated by the stations modelled by a
Poisson Process.
254
259
 TABLE II. PARAMETERS USED IN THE ANALYSIS
‫ܹܥ‬௠௜௡ 32 Retry limit ሺ݉ሻ 7 therefore the delay for access to the wireless channel
‫ܹܥ‬௠௔௫  1024 Basic Data Rate 1 ‫ݏ݌ܾܯ‬
Slot time 20 ߤ‫ݏ‬ Channel Data Rate 11 ‫ݏ݌ܾܯ‬
DIFS 50 ߤ‫ݏ‬ Propagation delay 2 ߤ‫ݏ‬
SIFS 10 ߤ‫ݏ‬ ACK Frame Payload 112 ܾ݅‫ݏݐ‬
MAC hdr 224 bits PHY Header 192 ܾ݅‫ݏݐ‬
B. Performance Evaluation
Figure (3) depicts the throughput, end-to-end delay and
frame loss probability of the WLAN under study with
different traffic models. The figures reveal a good match
between the analytical and the simulation results, therefore
proving that the developed model is accurate. There are
minor discrepancies between the model and the simulation
results at some stages which are mainly due to the
approximations taken into consideration to make the model
tractable. An instance of the approximations is the
assumption made that the collision probability is the same
at all times regardless of the backoff stage of the DCF
protocol. However the extensive comparisons made
between the analytical results and those gained from
simulation prove that the model has an acceptable
accuracy. It is evident from the results that bursty traffic
models have significant impacts on the network
performance. At lower rates, the throughput (Figure 3.a) of
the WLAN with different models is similar, but as soon as
the load reaches around 0.2 Mbps differences start to
appear. For the Poisson and 3-state MMPP the results
remain close up to when the load reaches 0.3 Mbps. After
that, with continuous increase in the load, the network
reaches saturation point for 3-state MMPP while in the
Poisson model the throughput continues to increase until
reaching a steady value at 0.34 Mbps.
In both the Poisson model and the 3-state MMPP, the
WLAN reaches saturation point when the load reaches 0.4
Mbps, whereas in the 3-state BMAP with batch sizes of 3
and 5, the network becomes saturated at an earlier stage,
when the load is around 0.2 Mbps.
Overall the maximum throughput of the network
during saturation is higher for Poisson, 0.34 Mbps, than
the other three traffic models. After Poisson (Figure 3.a),
stands the 3-state MMPP with 0.29 Mbps, then 3-state
BMAP with batch size of maximum three with 0.17 Mbps
and finally is the 3-state BMAP with maximum batch size
of five, with 0.12 Mbps. The same analysis stands for the
frame loss probability (Figure 3.b) and the end-to-end
delay (Figure 3.c). The network has the highest loss when
the traffic is modelled using the three-state BMAP with
batch size of maximum five, even higher than three-state
BMAP with batch size of maximum three. The 3-state
MMPP has a lower loss probability than the two BMAPs
but still has a higher loss rate compared to the Poisson
model. The results are logical as BMAP with maximum
batch size of five generates more burstiness, whereas the
Poisson process and the MMPP process generate one
packet at each instance. The end-to-end delay comparison
(Figure 3.c) shows that the delay in the WLANs with
highly bursty traffic reaches a high level at even lower
loads. When the traffic is modelled using Poisson or
MMPP, the network has a period of very low end-to-end
delay before a sharp increase which then stays nearly the
same during the saturation period. The reason is that with
highly bursty traffic models, even at low rates, there are
255
260
more packets competing for the shared media and
increases. Overall it can be seen from the results that the
QoS measures of stations under bursty traffic are similar to
those under Poisson traffic when the network operates at
low traffic loads (less than 0.2 Mbps). This is because the
collision probability at lower rates is very small hence the
low frame loss probability and queueing delays. However
from the figures it is very obvious that as the load
increases, the network performance and the QoS measures
start to drift apart rapidly under bursty traffic. This is an
obvious proof to the fact that the traffic generated by
multimedia applications is highly bursty and correlated,
and the conventional processes such as Poisson are no
longer adequate for modelling current network traffic.
VI. CONCLUSION
In this paper, an analytical model of 802.11-based
WLAN under video traffic is presented. BMAP is used to
model the video traffic in order to capture its bursty and
correlated characteristics. To obtain the queueing
dynamics, the transmission queue at each station has been
modelled as a BMAP/M/1/N queueing system. The
analytical model can derive the QoS performance metrics
including throughput, end-to-end delay and loss
probability. The accuracy of the analytical model is
validated through NS2 simulation experiments. The results
have shown the effects of BMAP traffic on the degradation
of QoS performance of 802.11-based WLANs, compared
to traffics modelled using Poisson and MMPP. This
highlights the fact that burstiness and correlation of video
traffic greatly deteriorates the QoS performance of
WLANs.
REFERENCES
[1] http://www.cisco.com/c/en/us/solutions/collateral/service-
provider/visual-networking-index-vni/mobile-white-paper-c11-
520862.html. (2016, may). Cisco Visual Networking Index: Global
Mobile Data Traffic Forecast Update, 2015–2020 White Paper.
[2] J. Beran, R. Sherman, M. S. Taqqu, and W. Willinger, "Long-range
dependence in variable-bit-rate video traffic," IEEE Transactions
on Communications, vol. 43, pp. 1566-1579, 1995.
[3] O. Tickoo and B. Sikdar, "On the impact of IEEE 802.11 MAC on
traffic characteristics," IEEE Journal on Selected Areas in
Communications, vol. 21, pp. 189-203, 2003.
[4] X. Sun, H. Cui, R. Liu, J. Chen, and Y. Liu, "Multistep ahead
prediction for real-time VBR video traffic using deterministic echo
state network," in 2012 IEEE 2nd International Conference on
Cloud Computing and Intelligence Systems, 2012, pp. 928-931.
[5] O. C. Kwon, Y. Go, and H. Song, "An Energy-Efficient
Multimedia Streaming Transport Protocol Over Heterogeneous
Wireless Networks," IEEE Transactions on Vehicular Technology,
vol. 65, pp. 6518-6531, 2016.
[6] K. Park and W. Willinger, "Self-Similar Network Traffic: An
Overview," in Self-Similar Network Traffic and Performance
Evaluation, ed: John Wiley & Sons, Inc., 2002, pp. 1-38.
[7] G. V. d. Auwera, P. T. David, and M. Reisslein, "Traffic
characteristics of H.264/AVC variable bit rate video," IEEE
Communications Magazine, vol. 46, pp. 164-174, 2008.
[8] D. M. Lucantoni, "The BMAP/G/1 QUEUE: A Tutorial,"
presented at the Performance Evaluation of Computer and
Communication Systems, Joint Tutorial Papers of Performance '93
and Sigmetrics '93, 1993.
[9] A. Chydzinski and R. Winiarczyk, "On the blocking probability in
batch Markovian arrival queues," Microprocess. Microsyst., vol.
32, pp. 45-52, 2008.
[10] E. Charfi, L. Chaari, and L. Kamoun, "PHY/MAC Enhancements [29] W. Haitao, P. Yong, L. Keping, C. Shiduan, and M. Jian,
and QoS Mechanisms for Very High Throughput WLANs: A "Performance of reliable transport protocol over IEEE 802.11
Survey," IEEE Communications Surveys & Tutorials, vol. 15, pp. wireless LAN: analysis and enhancement," in INFOCOM 2002.
1714-1735, 2013. Twenty-First Annual Joint Conference of the IEEE Computer and
[11] G. Min, J. Hu, and M. E. Woodward, "Performance Modelling and Communications Societies. Proceedings. IEEE, 2002, pp. 599-607
Analysis of the TXOP Scheme in Wireless Multimedia Networks vol.2.
with Heterogeneous Stations," IEEE Transactions on Wireless [30] http://trace.eas.asu.edu/videotraces2/4k/tos_1920_265_B0/20/.
Communications, vol. 10, pp. 4130-4139, 2011.
[12] T. Hayajneh and G. Al-Mashaqbeh, "Multimedia traffic over
WLANs: QoS support and performance evaluation," in Information
and Communication Systems (ICICS), 2014 5th International
Conference on, 2014, pp. 1-6.
[13] A. Abdrabou and W. Zhuang, "Service Time Approximation in
IEEE 802.11 Single-Hop Ad Hoc Networks," IEEE Transactions
on Wireless Communications, vol. 7, pp. 305-313, 2008.
[14] X. Zhang, "A New Method for Analyzing Nonsaturated IEEE
802.11 DCF Networks," IEEE Wireless Communications Letters,
vol. 2, pp. 243-246, 2013.
[15] W. Fischer and K. Meier-Hellstern, "The Markov-modulated
Poisson process (MMPP) cookbook," Performance Evaluation, vol.
18, pp. 149-171, 1993/09/01 1993.
[16] A. Abdrabou and W. Zhuang, "Stochastic delay guarantees and
statistical call admission control for IEEE 802.11 single-hop ad hoc
networks," IEEE Transactions on Wireless Communications, vol.
7, pp. 3972-3981, 2008.
[17] Y. He, R. Yuan, and W. Gong, "Modeling Power Saving Protocols
for Multicast Services in 802.11 Wireless LANs," IEEE
Transactions on Mobile Computing, vol. 9, pp. 657-671, 2010.
[18] H. Zhou, B. Li, Q. Qu, and Z. Yan, "An analytical model for
Quality of Experience of HTTP video streaming over wireless ad
hoc networks," in Signal Processing, Communication and
Computing (ICSPCC), 2013 IEEE International Conference on,
2013, pp. 1-5.
[19] Y. Gao, C. W. Tan, Y. Huang, Z. Zeng, and P. R. Kumar,
"Characterization and Optimization of Delay Guarantees for Real-
Time Multimedia Traffic Flows in IEEE 802.11 WLANs," IEEE
Transactions on Mobile Computing, vol. 15, pp. 1090-1104, 2016.
[20] "IEEE Standard for Information Technology- Telecommunications
and Information Exchange Between Systems-Local and
Metropolitan Area Networks-Specific Requirements-Part 11:
Wireless LAN Medium Access Control (MAC) and Physical Layer
(PHY) Specifications," IEEE Std 802.11-1997, pp. i-445, 1997.
[21] S. H. Nguyen, H. L. Vu, and L. L. H. Andrew, "Packet Size
Variability Affects Collisions and Energy Efficiency in WLANs,"
in 2010 IEEE Wireless Communication and Networking
Conference, 2010, pp. 1-6.
[22] D. M. Lucantoni, "New results on the single server queue with a
batch Markovian arrival process," Communications in Statistics.
Stochastic Models, vol. 7, pp. 1-46, 1991.
[23] A. N. Dudin, A. A. Shaban, and V. I. Klimenok, "Analysis of a
queue in the BMAP/G/1/N system," International Journal of
Simulation, vol. 6, pp. 13-23, 2005.
[24] A. N. Dudin, V. M. Vishnevsky, and J. V. Sinjugina, "Analysis of
the BMAP/G/1 queue with gated service and adaptive vacations
duration," Telecommunication Systems, vol. 61, pp. 403-415,
2016.
[25] A. Klemm, C. Lindemann, and M. Lohmann, "Modeling IP traffic
using the batch Markovian arrival process," Performance
Evaluation, vol. 54, pp. 149-173, 2003.
[26] P. Salvador, A. Pacheco, and R. Valadas, "Modeling IP traffic:
joint characterization of packet arrivals and packet sizes using
BMAPs," Computer Networks, vol. 44, pp. 335-352, 2004.
[27] G. Bianchi, "Performance analysis of the IEEE 802.11 distributed
Figure 3: Comparison of the results from analytical models and
coordination function," IEEE Journal on Selected Areas in
simulations for WLANs with stations generating traffic using Poisson,
Communications, vol. 18, pp. 535-547, 2000.
MMPP and BMAP.
[28] D. Malone, K. Duffy, and D. Leith, "Modeling the 802.11
Distributed Coordination Function in Nonsaturated Heterogeneous
Conditions," IEEE/ACM Transactions on Networking, vol. 15, pp.
159-172, 2007.

256
261