International
Journal
of
Electrical
and
Computer
Engineering
(IJECE)
V
ol.
10,
No.
5,
O
c
tober
2020,
pp.
5420
5429
ISSN:
2088-8708,
DOI:
10.11591/ijece.v10i5.pp5420-5429
r
5420
Influence
of
v
arious
application
types
on
the
perf
ormance
of
L
TE
mobile
netw
orks
Adel
Agamy,
Ahmed
M.
Mohamed
Electrical
Engineering
Department,
F
aculty
of
Engineering,
Asw
an
Uni
v
ersity
,
Egypt
Article
Inf
o
Article
history:
Recei
v
ed
Jan
16,
2020
Re
vised
Apr
16,
2020
Accepted
Apr
29,
2020
K
eyw
ords:
L
TE
Queuing
theory
T
raf
fic
modeling
W
ireless
netw
ork
ABSTRA
CT
Modern
mobile
internet
netw
orks
are
becoming
hea
vier
and
denser
.
Also
it
is
not
re
g-
ularly
planned,
and
becoming
more
heterogeneous.
The
e
xplosi
v
e
gro
wth
in
the
usage
of
smartphones
poses
numerous
challenges
for
L
TE
cellular
netw
orks
design
and
im-
plementation.
The
performance
of
L
TE
netw
orks
with
b
ursty
and
self-similar
traf
fic
has
become
a
major
challenge.
Accurate
modeling
of
the
data
generated
by
e
ach
con-
nected
wireless
de
vice
is
important
for
properly
in
v
estig
ating
the
performance
of
L
TE
netw
orks.
This
paper
presents
a
mathematical
model
for
L
TE
netw
orks
using
queuing
theory
considering
the
influence
of
v
arious
application
types.
Usi
ng
sporadic
source
traf
fic
feeding
to
the
queue
of
the
e
v
olv
ed
nodeB
and
with
the
e
xponential
service
time
assumption,
we
construct
a
queuing
model
to
estimate
the
performance
of
L
TE
netw
orks.
W
e
use
the
performance
model
presented
in
this
paper
to
study
the
influence
of
v
arious
application
cate
gories
on
the
performance
of
L
TE
cellular
netw
orks.
Also
we
v
alidate
our
model
with
simulation
using
NS3
simulator
with
dif
ferent
scenarios.
Copyright
c
2020
Insitute
of
Advanced
Engineeering
and
Science
.
All
rights
r
eserved.
Corresponding
A
uthor:
Adel
Ag
amy
,
Electrical
Engineering
Department,
F
aculty
of
Engineering,
Asw
an
Uni
v
ersity
,
Asw
an,
81542,
Egypt
Email:
a.f.ag
amy@aswu.edu.e
g
1.
INTR
ODUCTION
The
incredible
gro
wth
in
the
number
of
wireless
de
vices
such
as
smart-phones,
tablets
and
Inter
-
net
of
Thing
(IoT)
in
addition
to
the
f
ast
de
v
elopment
of
media
streaming
applications,
IPTV
,
telemedicine
and
Internet
g
aming
ha
v
e
led
to
a
significant
chall
enge
to
the
design
and
deplo
yment
of
cellular
technology
.
The
mobile
netw
orks
specifically
the
L
TE
netw
orks
are
used
only
for
accommodating
v
oice
and
video
calls
traf
fic
which
are
considered
real
time
application.
Also
no
w
adays
the
mobile
netw
orks
are
used
to
transfer
non-real
time
data
(Email,
ftp,
..etc).
Each
application
type
demands
to
maintain
a
certain
le
v
el
of
quality
(throughput,
delay
..etc)
during
his
sojourn
time
through
the
L
TE
netw
ork
[1,
2,
3,
4].
In
v
estig
ating
and
ana-
lyzing
the
distrib
ution
of
data
generated
by
each
de
vice
in
L
TE
netw
ork
should
be
the
most
important
f
actor
to
estimate
the
quality
requirements
and
capabilities
of
L
TE
netw
orks.
In
this
research
we
in
v
estig
ate
and
analyze
(by
analytical
modeling
and
simulation)
the
influence
of
v
arious
application
types
on
the
L
TE
net
w
ork
perfor
-
mance.
Also
we
v
alidate
our
model
through
real
netw
ork
simulator
NS3
with
v
arious
scenarios.
Our
paper
is
or
g
anized
as
follo
ws,
pre
vious
studies
related
to
the
paper
topic
is
presented
in
section
II.
The
L
TE
netw
ork
system
model
and
its
pa
rameters
are
presented
in
section
III.
Section
IV
sho
ws
the
L
TE
netw
ork
performance
estimation
using
anal
ytical
analysis
with
netw
ork
performance
metrics
such
as
end
to
end
pack
et
delay
and
blockage
probability
.
Section
V
sho
ws
the
performance
beha
vior
of
the
L
TE
netw
ork
using
NS3
simulator
with
dif
ferent
operation
scenarios.
In
section
VII
we
conclude
our
w
ork.
J
ournal
homepage:
http://ijece
.iaescor
e
.com/inde
x.php/IJECE
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Elec
&
Comp
Eng
ISSN:
2088-8708
r
5421
2.
RELA
TED
W
ORK
Sajid
et
all
in
[5]
de
v
eloped
a
model
using
tw
o
dif
ferent
distrib
utions.
F
or
Constant
bit
Rate
(CBR)
traf
fic
(Ex.
V
OIP
traf
fic)
the
y
used
e
xponential
distrib
ution.
T
o
model
V
ariable
Bit
Rate
(VBR)
traf
fic
(video
streaming)
the
y
used
fractional
bro
wnie
motion
(FBM)
traf
fic
with
hea
vy-tailed
W
eib
ull
distrib
ution
for
b
uf
fer
occupanc
y
[6].
The
authors
in
[7]
in
v
estig
ated
the
performance
of
cogniti
v
e
radio
links
subject
to
recurrent
f
ailures
and
interruptions.
The
y
studied
the
performance
wit
h
single
and
multiple
channels.
Using
the
queuing
model
M/M/1,
the
y
considered
the
service
interruption
from
primary
users,
also
the
y
used
the
pricing
polic
y
to
char
ge
each
secondary
user
in
the
queue.
Authors
in
[8]
proposed
a
queueing
model
with
four
dif
ferent
priority
queueing
disciplines
to
apply
dynamic
optimization.
The
y
considered
a
dynamic
priority
queueing
discipline
to
optimize
a
joint
performance
utility
function
on
tw
o
classes
of
cogniti
v
e
radio.
The
performance
of
v
ehicular
netw
ork
communication
using
cellular
L
TE
netw
ork
using
queueing
models
is
presented
in
[9,
10,
11,
12,
13].
F
or
e
xample
the
author
in
[11]
de
v
eloped
an
analytical
model
describ-
ing
the
performance
of
periodic
broadcast
in
V
ehicular
Ad-hoc
Netw
orks
(V
ANET)
in
terms
of
pa
ck
et
collision
probability
and
a
v
erage
pack
et
delay
using
M/M/
1
queuing
model.
Analytical
models
to
e
v
aluate
t
he
queue
length
beha
vior
at
the
intersection
points
as
a
function
of
the
percentage
of
v
ehicles
are
presented
in
[12].
The
authors
in
[13]
used
deterministic
arri
v
al
process
and
the
queueing
model
D/M/1
for
studying
the
performance
of
periodic
broadcast
in
V
ANETs
using
metrics
such
as
the
pack
et
collision
probability
and
a
v
erage
pack
et
delay
.
The
authors
in
[14]
used
the
queuing
theory
to
de
v
elop
a
model
for
cellular
L
TE
wireless
netw
orks.
The
y
assumed
that
the
cellular
L
TE
netw
ork
is
serving
v
ariable
bit-rate
calls.
The
authors
in
[15]
e
v
aluated
the
polling
beha
vior
on
a
MA
C
for
cellular
netw
ork
analytically
considering
the
pack
et
delay
,
b
uf
fer
o
v
erflo
w
rates
and
ener
gy
consumption.
Also
in
their
w
ork
in
[16]
the
y
introduced
analytical
models
to
characterize
the
delay
for
multicast
transmission
o
v
er
a
communication
channel
model.
In
[17],
authors
presented
a
mix
ed
queueing
netw
ork
models
of
se
v
eral
mobility
users
at
numerous
access
points
to
accurately
predicting
the
number
of
netw
ork-le
v
el
performance
and
user
-le
v
el
performance
in
a
wireless
netw
ork.
In
[18]
Scott
et
all
introduced
a
model
for
V
ehicular
W
ireless
Channel
Communication,
the
y
modeled
the
L
TE
system
channel
with
M/M/m
Queueing
model.
The
y
modeled
the
L
TE
wireless
communication
channel
with
M/M/m
queueing
model
with
infinite
number
of
resources
channel
m.
Also
the
y
assumed
that
each
v
ehicle
generates
traf
fic
with
Poisson
process
of
e
xponential
inter
arri
v
al
time
distrib
ution.
The
y
used
the
First
Come
First
Send
(FCFS)
queueing
discipline
to
handle
traf
fic
in
the
queue.
The
y
e
v
aluated
the
probability
that
channels
being
b
usy
,
also
de
v
eloped
the
e
xpected
w
ait
ing
times
and
the
e
xpected
number
of
channel
switches.
F
o
wler
et
all
in
[19]
described
the
beha
vior
of
the
L
TE
video
call
(e
x.
Sk
ype
video
call)
and
video
streaming
traf
fic
in
heterogeneous
real
en
viron-
ment
using
g
aussian
mixture
model.
The
y
deri
v
ed
a
semi-Mark
o
v
model
with
six
states
for
video
call
and
the
y
deri
v
ed
a
semi-Mark
o
v
model
with
fifteen
states
to
fit
the
statistics
of
composite
L
TE
video
measurements.
Najem
et
all
in
[20]
used
the
Disjoint
Queue
Scheduler
(DQS)
for
the
L
TE-A
heterogeneous
netw
ork
deplo
yment
of
a
macro-cell
and
a
v
ariable
number
of
picocells.
The
y
e
v
aluated
and
compared
the
Quality
of
Service
(QoS)
of
the
user
performance
using
the
DQS
based
on
dif
ferent
techniques.
The
y
e
v
aluated
the
QoS
using
the
a
v
erage
subscriber’
s
metrics
t
hroug
hput
,
P
ack
et
Loss
Rate
(PLR),
and
a
v
erage
pack
et
delay
.
The
y
e
xperimentally
e
v
aluated
the
performance
of
the
DQS
and
its
ef
fect
on
the
user
qual
ity
of
service.
Naumo
v
et
all
in
[21]
st
udied
the
performance
of
L
TE
cellular
netw
orks
using
queueing
models
with
limited
resources.
The
y
de
v
eloped
a
mathematical
model
for
resource
sharing
in
L
TE
cellular
netw
ork
using
multi-serv
er
queueing
model.
The
y
assumed
that
users
arri
v
e
in
the
system
as
independent
Poisson
flo
ws.
The
service
time
for
each
arri
v
al
is
modeled
with
e
xponential
distrib
ution.
The
authors
considered
a
multi-serv
er
queuing
model
and
the
y
used
semi-mark
o
v
chains
to
deri
v
e
the
stationary
probabilities
for
the
L
TE
cellular
netw
ork
with
single
L
TE
netw
orks
serving
users
using
video
conference
call.
Polag
ang
a
et
al
l
in
[22]
e
xplored
the
Self-Similarity
property
of
L
TE
and
L
TE-adv
ance
cellular
netw
orks.
The
y
sho
wed
that
the
selfsimilarity
characteristics
of
L
TE
and
L
TE-adv
anced
cellular
netw
orks
traf
fic,
also
the
y
e
v
aluated
and
compared
selfsimilari
ty
de
gree
for
both
netw
orks
and
compared
user
traf
fic
with
traditional
v
oice
traf
fic.
The
y
summarized
from
dif
ferent
data
sets
that
the
arri
v
al
pattern
of
the
user
in
real
L
TE
netw
orks
follo
ws
Poisson
proces
s.
Also,
the
y
found
that
the
inter
arri
v
al
time
follo
ws
the
Exponential
distrib
ution.
Based
on
our
kno
wledge
there
aren’
t
an
y
pre
vious
studies
that
used
sporadic
and
hea
vy
tail
characteristics
of
the
L
TE
cellular
traf
fic
in
one
model.
W
ith
sporadic
input
traf
fic
feed
to
the
eNodeB
queue,
we
deri
v
ed
performance
metrics
such
as
mean
pack
et
delay
and
blocking
probability
analytically
and
v
erified
that
using
the
NS3
simulator
under
v
arious
b
urst
v
alues
(v
arious
application)
with
fix
ed
utilization.
W
e
choose
the
po
wer
tail
distrib
ution
with
v
arious
truncated
tail
v
alues
to
represent
the
number
of
pack
ets
during
a
request.
The
reliability
function
of
the
po
wer
tail
distrib
ution
used
is:
Influence
of
various
application
types
on
the
performance
of
...
(Adel
Agamy)
Evaluation Warning : The document was created with Spire.PDF for Python.
5422
r
ISSN:
2088-8708
R
(
x
)
:=
1
1
T
T
1
X
j
=0
j
exp
x
j
(1)
T
=
1
refers
to
the
e
xponential
and
lar
ge
T
for
highly
tail
properties[23].
3.
L
TE
NETW
ORK
SYSTEM
MODEL
The
topology
used
for
L
TE
netw
orks
is
illustra
ted
in
Figure
1.
It
consists
of
a
single
e
v
olv
ed
nodeB
and
a
set
of
N
wireless
de
vices
that
access
the
L
TE
netw
ork.
All
of
the
wireless
de
vices
use
the
L
TE
wire-
less
access
technology
for
do
wnloading
data
(using
do
wnlink)
and
to
request
and
upload
data
(using
uplink).
W
e
focus
on
do
wnlink
communication
and
assume
tha
t
these
de
vices
use
v
arious
mobile
applications.
Also
the
corresponding
request
traf
fic
in
the
L
TE
do
wnlink
access
netw
ork
ha
v
e
v
arious
properties
due
to
the
dif
ferent
mobile
applications
used.
The
number
of
pack
ets
during
a
user
request
is
a
random
v
ariable
and
depends
on
the
mobile
application
used,
so
the
distrib
ution
of
do
wnlink
entrance
process
will
be
dif
ferent
wi
th
each
mobile
ap-
plication
type.
Ahmed
et
al
in
[24]
compared
v
arious
traf
fic
model
schemes
to
model
the
Int
ernet
data
and
ho
w
each
model
can
capture
beha
vior
of
the
real
application
traf
fic.
the
y
sho
wed
that
the
b
ursty
traf
fic
introduced
in
[23]
is
the
best
distrib
ution
to
model
v
arious
application
types
due
to
its
v
arious
parameters.
The
model
can
represent
the
b
ursty
traf
fic
(with
the
idle
and
acti
v
e
periods)
as
in
Figure
2
and
also
produces
the
self-similarity
property
which
presents
man
y
modes
of
use
(continuous
flo
w
,
Bulk
arri
v
al,
Poisson
arri
v
al
etc)
[17,
19].
Figure
1.
L
TE
netw
ork
topology
Figure
2.
ON/OFF
Model
The
essential
parameters
of
the
traf
fic
model
used
are
as
introduced
in
[24]:
K:=
the
a
v
erage
rate
of
pack
ets
arri
v
al
of
connected
L
TE
de
vices.
:=
the
aggre
g
ated
rates
that
produced
by
the
N-L
TE
de
vices
where
=KN.
n
p
:=
The
a
v
erage
number
of
pack
ets
produced
during
in
the
b
urst.
Int
J
Elec
&
Comp
Eng,
V
ol.
10,
No.
5,
O
c
tober
2020
:
5420
–
5429
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Elec
&
Comp
Eng
ISSN:
2088-8708
r
5423
p
:=
the
maximum
rate
that
a
de
vice
can
send
during
an
acti
v
e
period.
O
N
:=
n
p
/
p
=
a
v
erage
time
for
an
acti
v
e
period.
O
F
F
:=
a
v
erage
idle
time
between
tw
o
b
urst
sending
periods.
As
sho
wn
in
Figure
3
the
L
TE
netw
ork
consists
of
the
e
v
olv
ed
pack
et
node
(eNodeB)
which
is
respon-
sible
for
allocating
resources
to
the
connected
de
vices
and
the
service
g
ate
w
ay
(S-GW)
and
pack
et
g
ate
w
ay
(P-GW)
which
connects
to
the
eNodeB
through
MME
unit
and
from
MME
to
the
internet.
Also
the
connected
de
vices
(smart-phones)
with
b
ursty
traf
fic
model
for
the
multi-applications.
W
e
assume
that
all
de
vices
are
constant
(so
no
hando
v
er
to
other
neighbor
cells).
W
e
also
assume
that
the
resources
allocation
in
the
eNodeB
uses
the
round
robin
scheduling
so
that
the
resources
are
di
vided
equally
among
the
users.
W
e
use
the
whole
single
L
TE
cell
topology
as
in
Figure
3
with
Nb
urst/M/1
queuing
model.
No
w
we
outline
the
single
cell
L
TE
traf
fic
model
using
sporadic
traf
fic
model:
:=
a
v
erage
service
rate
of
an
eNodeB.
U:=
/
=
Load
utilization
of
the
access
eNodeB.
Also
we
can
control
on
the
sporadic
type
with
parameter
”b”,
which
can
get
as
in
[23]
from
the
follo
wing:
b
=
1
K
R
p
=
1
nR
p
(2)
Figure
3.
L
TE
Netw
ork
Model
4.
L
TE
AN
AL
YTICAL
PERFORMANCE
MODEL
4.1.
P
erf
ormance
metrics
The
analysis
of
a
single
cell
L
TE
acces
s
netw
ork
performance
with
sporadic
can
be
deri
v
ed
for
al
l
v
arious
distrib
utions
using
matrix-e
xponential
approach
[25].
As
we
mentioned
abo
v
e
that
there
man
y
modes
of
use
to
our
model.
The
first
mode
of
use
for
the
model
where
the
idle
period
approaches
zero
which
lead
to
continuous
flo
w
(no
b
ursts
and
b=0)
so
the
model
can
be
reduced
to
Poisson
arri
v
al
(
M
/
M
/l
queueing
model)
and
hence
the
delay
can
be
deri
v
ed
from
the
follo
wing
equation
as
in
[25]:
Mean
Delay(b=
0)=((1/
)/(1-U))
Where
U=
/
.
The
second
mode
of
use
occurs
when
the
acti
v
e
tim
e
approaches
zero,
in
this
case
the
pack
ets
arri
v
e
as
a
b
ulk
arri
v
al
where
”b=1”.
So
the
delay
can
be
calculated
as
in
[25]
from
equati
on:
M
ean
De-
lay(b=1)=(D
(1/
)/(1-U)),..
where
D
=
E
(
L
(
L
+1)
2
)
E
(
L
)
.
where
the
best
performance
of
the
cellular
netw
ork
is
in
the
fir
st
mode
of
use
where
”b=0”,
the
w
orst
performance
is
in
the
second
mode
of
use
when
”b=1”.
Through
the
analytical
analysis
we
use
fix
ed
load
utilization
ˇ
U
at
eNodeB,
while
we
in
v
estig
ate
the
influence
of
b
ursty
de
gree
(dif
ferent
mobile
application).
The
eNodeB
load
utilization
is
fix
ed
while
the
size
of
the
b
ursts
increases
or
decreases
according
to
the
de
gree
of
b
urstiness
”b”
to
capture
the
influence
of
dif
ferent
applications
(each
v
alue
of
b
represents
a
traf
fic
type).
The
queueing
model
can
be
represented
with
matrix
e
xponential
approach
of
multiple
application
types.
The
steady
state
solution
for
the
system
can
be
deri
v
ed
as
in
[23]
so
we
can
get:
The
end
to
end
pack
et
delay
is
calculated
using
little
0
s
formula:
D
E
LAY
=
1
K
R
(
I
R
)
1
::::::
(3)
Influence
of
various
application
types
on
the
performance
of
...
(Adel
Agamy)
Evaluation Warning : The document was created with Spire.PDF for Python.
5424
r
ISSN:
2088-8708
Also
we
can
get
the
block
probability
as
follo
ws:
B
l
ock
P
r
oabil
ity
=
1
K
(
R
B
L
)
:
(4)
Where
the
matrix
R
can
be
calculated
by
solving
the
system
as
Quasi-Birth-Death
Process,
K
is
the
a
v
erage
arri
v
al
rate,
I
is
the
identity
matrix
and
is
unity
v
ector
.
4.2.
Model
analysis
The
first
scenario
sho
ws
the
topology
of
the
L
TE
netw
ork
as
in
Figure
2.
W
e
assume
that
the
load
arri
v
es
to
the
eNodeB
as
a
single
flo
w
(N=1
in
the
model
in
Figure
3).
In
all
scenarios
we
set
the
idle
period
to
the
e
xponential
distrib
ution
where
during
acti
v
e
time
ea
ch
de
vice
produce
a
flo
w
with
a
random
size
(Po
wer
tail
distrib
ution
with
v
arious
tail
v
alues)
to
represent
the
qualitati
v
e
statistical
manner
of
v
arious
mobile
applications
traf
fic.
Setting
truncated
tail
distrib
ution
to
”1”
refers
to
e
xponential
distrib
ution.
First
Figure
4
sho
ws
the
relation
between
the
end
to
end
pack
et
delay
and
the
b
urstiness
parameter
”b”.
W
e
notice
that
the
delay
increases
significantly
with
the
increase
of
the
b
urstiness
parameter
”b”,
also
the
delay
jumps
to
a
lar
ge
v
alue
at
point
(b=0.1)
and
then
starts
to
increase
gradually
.
More
significantly
the
pack
et
delay
of
the
applications
which
follo
w
distrib
ution
with
tail
equal
to
”28”
is
almost
twice
the
delay
of
the
applications
that
follo
w
the
e
xponential
distrib
ution.
So,
assuming
that
all
applications
will
follo
w
the
e
xponential
distrib
ution
as
a
service
time
distrib
ution
will
lead
to
an
o
v
er
estimate
of
the
L
TE
netw
ork
capa-
bilities.
These
estimates
in
Figure
5
are
based
on
only
a
single
user
in
the
netw
ork
so
the
ef
fect
of
contention
is
minimum.
So
e
v
en
with
no
contention,
t
he
application
type
has
a
significant
impact
on
the
pack
et
delay
of
the
L
TE
netw
orks.
Figure
5
sho
ws
the
beha
vior
of
blocking
probability
v
ersus
the
b
urstiness
parameter
”b”.
The
figure
sho
ws
clearly
ho
w
the
application
type
can
af
fect
the
user
block
probability
on
the
L
TE
netw
orks.
Clearly
applications
with
high
v
ariability
(long
tails)
starts
to
ha
v
e
significant
blockage
probability
earlier
than
applications
with
small
v
ariability
(e
xponential
distrib
utions).
F
or
small
v
alues
of
”b”
there
i
sn’
t
a
significant
dif
ference
between
v
arious
v
alues
of
blockage
probability
for
truncated
tail
and
the
e
xponential
distrib
ution,
as
b
urstines
s
parameter
”b”
increases,
the
g
ap
between
the
e
xponential
and
other
high
v
ariability
distrib
utions
increases.
By
taking
into
account
(2)
and
(3),
we
can
find
out
that
the
jump
happen
if
R
p
>
,where
b
>
”l
-
U”
at
this
moment.
Figure
4.
Single
user
mean
delay
with
dif
ferent
b
urst
Figure
5.
Single
block
probability
Int
J
Elec
&
Comp
Eng,
V
ol.
10,
No.
5,
O
c
tober
2020
:
5420
–
5429
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Elec
&
Comp
Eng
ISSN:
2088-8708
r
5425
Our
ne
xt
scenario
as
the
netw
ork
topology
in
Figure
3.
W
e
assume
that
the
arri
v
al
at
the
eNodeB
comes
from
tw
o
dif
ferent
sources
(N=2),
each
one
with
ON
periods
that
follo
w
po
wer
t
ail
distrib
ution
and
e
xponential
OFF
period
distrib
ution.
The
inte
gration
of
”N”
multiple
e
xponential
distrib
utions
with
rate
”r”
is
Nr
e
xponential
distrib
ution
[25].
Hence,
we
can
use
the
(
N-b
urs
t/M/1)
queueing
model.
F
or
small
v
alues
of
”b”,
the
model
reduces
to
a
continuous
arri
v
al
flo
w
with
rate
NK
while
for
lar
ge
v
alues
of
”b”
the
model
approaches
the
b
ulk
arri
v
al
and
leads
to
w
orst
performance
on
the
L
TE
netw
ork.
T
o
get
a
f
air
comparison
with
the
single
load
source
scenario,
we
maintain
the
same
load
utilization
on
the
eNodeB.
The
results
in
Figures
6
and
7
are
consistent
with
the
single
flo
w
source
results.
The
dif
ference
in
the
tw
o
flo
w
sources
that
users
don’
t
suf
fer
from
a
lar
ge
jump
where
the
users
blo
w
up
points
need
long
b
ursts
to
occur
.
In
case
of
multiple
flo
w
”N=2”
we
ha
v
e
tw
o
jump
points
occur
as
in
Figure
6
for
the
delay
and
Figure
7
for
the
blocking
probability
.
Henc
e,
the
performance
of
the
delay
become
more
comple
x
for
lar
ge
v
alues
of
N.
W
e
can
estimate
the
jump
points
in
L
TE
netw
orks
with
a
b
ursty
traf
fic
source
from
the
follo
wing
equation:
b
=
1
K
R
p
=
1
N
R
p
(5)
Figure
6.
Multiple
users
mean
delay
with
dif
ferent
b
urst
Figure
7.
Multiple
users
block
probability
5.
L
TE
NETW
ORK
SIMULA
TION
Our
simulation
consists
of
a
single
zone
co
v
ered
by
an
L
TE
netw
ork
through
one
e
v
olv
ed
NodeB
(eNB)
as
in
Figure
2
and
L
TE
netw
ork
model
in
Figure
3.
The
specification
of
L
TE
technology
used
in
the
simulator
can
be
found
in
T
able
1.
The
smartphones
or
subscribers
are
uniformly
dis
trib
uted
in
the
zone
according
to
a
disc
around
the
eNodeB.
Each
de
vice
generates
load
for
am
ount
of
time
(acti
v
e
period)
and
idle
for
another
time
(OFF
period).
W
e
run
the
simulation
for
tw
o
scenarios
with
e
xponential
OFF
period
distrib
ution.
The
first
scenario
represents
the
normal
traf
fic
(e
xponential
for
act
i
v
e
period)
while
the
second
scenario
uses
the
hea
vy
tail
traf
fic
(P
areto
distrib
ution
for
acti
v
e
period).
W
e
run
the
simulator
for
v
arious
time
periods
and
then
calculate
the
a
v
erage
o
v
er
all
cases.
During
the
acti
v
e
period,
the
number
of
pack
ets
in
a
request
is
a
random
v
ariable
with
mean
of
1024
byte.
W
e
refer
to
the
number
of
pack
et
with
n
p
pack
ets
which
are
transmitted
with
a
constant
peak
rate
2
Influence
of
various
application
types
on
the
performance
of
...
(Adel
Agamy)
Evaluation Warning : The document was created with Spire.PDF for Python.
5426
r
ISSN:
2088-8708
Mb/s
during
an
acti
v
e
time.
Through
the
netw
ork
simulator
,
we
generate
the
dif
ferent
distrib
utions
from
the
follo
wing
function:
R
(
x
)
=
1
(1
+
x
M
(
1)
)
x
(6)
Where
R(x)
=
Pr(X
>
x):
a
reliability
function
of
the
random
v
ariable.
:..
a
shape
parameter
.
M
:..the
a
v
erage
of
the
distrib
ution,
(i.e.,
E(X)
=
M).
5.1.
Simulation
r
esult
analysis
Figure
8
sho
ws
the
aggre
g
ated
throughput
of
the
cell
(eNodeB)
for
both
the
small
v
ariability
distri-
b
ution
(e
xponential)
and
high
v
ariability
(P
areto)
traf
fic.
W
e
notice
from
the
Figure
8
that
the
total
aggre
g
ated
throughput
of
L
TE
netw
ork
for
hea
vy
traf
fic
(P
areto)
is
smaller
than
throughput
of
the
well
beha
v
ed
traf
fic.
The
result
from
the
Figure
8
confirms
with
what
we
got
in
the
analytical
model
and
the
cell
throughput
satu-
rated
at
approximately”33MB/s”
which
is
close
to
the
standard
throughput
set
in
T
able
1
[26].
Figure
9sho
ws
the
number
of
users
demand
o
v
er
time.
It
is
v
ery
clear
from
Figure
9
that
the
lar
ge
ne
g
ati
v
e
influence
of
hea
vy
tail
applications
on
the
performance
of
L
TE
netw
orks.
The
well-beha
v
ed
applications
netw
ork
can
serv
e
more
users
than
the
netw
ork
with
hea
vy
tail
applications.
The
Figures
10
and
11
sho
w
changes
of
pack
et
loss
per
-
centage
per
user
o
v
er
time.
The
figures
clearly
sho
w
that
a
lar
ge
part
of
users
lost
50
%
of
their
requests
in
P
areto
distrib
ution
while
the
number
decrease
to
13
%
on
the
e
xponential
di
strib
ution.
The
same
consistence
beha
vior
is
sho
wn
in
Figures
12
and
13
for
the
a
v
erage
delay
,
where
the
delay
for
hea
vy
tail
distrib
utions
reach
90
%
of
user
while
for
well-beha
v
ed
application
the
lar
ge
delay
ef
fects
70
%
of
users.
T
able
1.
L
TE
parameters
setting
Carrier
frequenc
y
..
..
2.6
GHZ
Bandwidth..
..
10
MHz
(50
RB)
Height
eNB..
..
25
m
T
ransmission
Po
wer
eNB
46
dBm
MIMO/SISO..
..SISO
eNB
noise
figure
5
dB
T
ransmission
po
wer
de
vices
24
dB
UE
noise
figure
7
dB
Scheduler
-
HARQ
Round
robin
Scheduler
-Y
es
T
ransmission
model
RLC
UM
Propag
ation
Loss
Model
T
w
oRayGround
Propag
ationLoss
Model
range
150
m
Antenna
model
Isotropic
Figure
8.
Throughput
Figure
9.
Demand
Int
J
Elec
&
Comp
Eng,
V
ol.
10,
No.
5,
O
c
tober
2020
:
5420
–
5429
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Elec
&
Comp
Eng
ISSN:
2088-8708
r
5427
Figure
10.
P
ack
et
loss
distrib
ution(EXP
Figure
11.
P
ack
et
loss
distrib
ution(P
areto)
Figure
12.
Delay
distrib
ution(EXP)
Figure
13.
Delay
distrib
ution(P
areto)
6.
CONCLUSIONS
W
e
in
v
estig
ated
the
influence
of
v
arious
types
of
mobile
application
traf
fic
on
the
performance
of
L
TE
mobile
netw
orks.
Our
research
focused
on
the
hea
vy-tailed
and
self-similar
statistical
characteristics
of
the
mobile
applications
and
its
ne
g
ati
v
e
ef
fect
on
the
L
TE
netw
ork
performance.
Using
the
(N-Burst/M/l)
Queuing
model
and
NS3
simulator
,
we
estimated
the
influence
of
application
types
on
L
TE
cellular
netw
ork
performance
beha
vior
.
Specifically
,
we
studied
L
TE
netw
ork
performance
metrics
such
as
pack
et
delay
,
block
probability
analytically
and
the
throughput,
pack
et
delay
,
pack
et
loss
and
user
demand
by
the
NS3
simulator
.
Our
future
w
ork
is
to
e
xtend
the
analytical
model
for
heterogeneous
netw
ork
taking
into
account
the
mobility
model
for
wireless
de
vices.
Also
to
e
v
aluate
more
complicated
scenarios
with
NS3
simulator
to
accurately
estimate
actual
netw
ork
traf
fic.
REFERENCES
[1]
Inde
x,
Cisco
V
isual
Netw
orking,
“Cisco
visual
netw
orking
inde
x:
Global
mobile
data
traf
fic
forecast
update
2015-2020
White
P
aper
,
”
Accessed
date,
2016.
Influence
of
various
application
types
on
the
performance
of
...
(Adel
Agamy)
Evaluation Warning : The document was created with Spire.PDF for Python.
5428
r
ISSN:
2088-8708
[2]
J.
W
u,
C.
Y
uen,
N.-M.
Cheung,
J.
Chen,
and
C.
W
.
Chen,
“Enabling
adapti
v
e
high-frame-rate
video
streaming
in
mobile
cloud
g
aming
applications,
”
IEEE
T
rans.
Circuits
Syst.
V
ideo
T
echn.,
v
ol.
25,
no.
12,
pp.
1988–2001,
2015.
[3]
N.
P
anw
ar
,
S.
Sharma,
and
A.
K.
Singh,
“
A
surv
e
y
on
5g:
The
ne
xt
generation
of
mobile
communication,
”
Ph
ysical
Communication,
v
ol.
18,
pp.
64–84,
2016.
[4]
A.
Gupta
and
R.
K.
Jha,
“
A
surv
e
y
of
5g
netw
ork:
Architecture
and
emer
ging
technologies,
”
IEEE
access,
v
ol.
3,
pp.
1206–1232,
2015.
[5]
M.
S.
Mushtaq,
S.
F
o
wl
er
,
B.
Augustin,
and
A.
Mel
louk,
“Qoe
in
5g
cloud
netw
orks
using
multimedia
services,
”
W
ireless
Communications
and
Netw
orking
Conference
(WCNC),
pp.
1–6,
2016.
[6]
I.
Norros,
“On
the
use
of
fractional
bro
wnian
motion
in
the
theory
of
connectionless
netw
orks,
”
IEEE
Journal
on
selected
areas
in
communications,
v
ol.
13,
no.
6,
pp.
953–962,
1995.
[7]
A.
Azarf
ar
,
J.-F
.
Frigo,
and
B.
Sanso,
“
Analysis
of
cogniti
v
e
radio
netw
orks
based
on
a
queueing
model
with
serv
er
interruptions,
”
IEEE
International
Conference
on
Communications,
pp.
1703–1708,
2012.
[8]
A.
Azarf
ar
,
J.-F
.
Frigon,
and
B.
Sans‘o,
“Dynamic
selection
of
priority
queueing
discipline
in
cogniti
v
e
radio
netw
orks,
”
V
ehicular
T
echnology
Conference
(VTC
F
all),
pp.
1–5,
2012.
[9]
Q.
Y
ang,
J.
Zheng,
and
L.
Shen,
“Modeling
and
performance
analysis
of
periodic
broadcast
in
v
ehicular
ad
hoc
netw
orks,
”
Global
T
elecommunications
Conference
(GLOBECOM
2011),
pp.
1–5,
2011.
[10]
S.
Y
ang,
C.
K.
Y
eo,
and
B.
S.
Lee,
“Predicti
v
e
scheduling
in
dri
v
e-thru
netw
orks
with
flo
w-le
v
el
dynamics
and
deadlines,
”
IEEE
International
Conference
on
Communications
(ICC),
pp.
1–5,
2011.
[11]
M.
J.
Khabbaz,
W
.
F
.
F
a
w
az,
and
C.
M.
Assi,
“
A
simple
free-flo
w
traf
fic
model
for
v
ehicular
intermit-
tently
connected
netw
orks,
”
IEEE
T
ransactions
on
Intelligent
T
ransportation
Systems,
v
ol.
13,
no.
3,
pp.
1312–1326,
2012.
[12]
G.
Comert
and
M.
Cetin,
“
Analyt
ical
e
v
aluati
o
n
of
the
error
in
queue
length
estimation
at
traf
fic
sig-
nals
from
probe
v
ehicle
data,
”
IEEE
T
ransactions
on
Intelligent
T
ransportation
Systems,
v
ol.
12,
no.
2,
pp.
563–573,
2011.
[13]
M.
J.
Khabbaz,
W
.
F
.
F
a
w
az,
and
C.
M.
Assi,
“Modeling
and
delay
analysis
of
intermittently
con-
nected
roadside
communication
netw
orks,
”
IEEE
T
ransactions
on
V
ehicular
T
echnology
,
v
ol.
61,
no.
6,
pp.
2698–2706,
2012.
[14]
M.
K.
Karray
and
M.
Jo
v
ano
vic,
“
A
queueing
theoretic
approach
to
the
dimensioning
of
wireless
cellular
netw
orks
serving
v
ariable-bi
t-rate
calls,
”
IEEE
T
ransactions
on
V
ehicular
T
echnology
,
v
ol.
62,
no.
6,
pp.
2713–2723,
2013.
[15]
H.
Y
ang
and
B.
Sikdar
,
“Queueing
analysis
of
polling
b
a
sed
wireless
mac
protocols
with
sleep-w
ak
e
c
ycles,
”
IEEE
T
ransactions
on
Communications,
v
ol.
60,
no.
9,
pp.
2427–2433,
2012.
[16]
B.
Shrader
and
A.
Ephremides,
“Queueing
delay
analysis
for
multicast
with
random
linear
coding,
”
IEEE
T
ransactions
on
Information
Theory
,
v
ol.
58,
no.
1,
pp.
421–429,
2012.
[17]
Y
.-C.
Chen,
J.
K
urose,
and
D.
T
o
wsle
y
,
“
A
mix
ed
queueing
netw
ork
model
of
mobility
in
a
campus
wireless
netw
ork,
”
Proceedings
IEEE
INFOCOM,
pp.
2656–2660,
2012.
[18]
S.
F
o
wler
,
C.
H.
Hall,
D.
Y
uan,
G.
Bara
vdish,
and
A.
Mellouk,
“
Analysis
of
v
ehicular
wireless
channel
communication
via
queueing
t
heory
model,
”
IEEE
International
Conference
on
Communications
(ICC),
pp.
1736–1741,
2014.
[19]
S.
F
o
wler
,
J.
Sarfraz,
M.
M.
Abbas,
and
V
.
Angelakis,
“Gaussian
s
emi-mark
o
v
model
based
on
real
video
multimedia
traf
fic,
”
IEEE
International
Conference
on
Communications
(ICC),
pp.
6971–6976,
2015.
[20]
N.
N.
Sirhan,
M.
Martınez-Ramon,
G.
L.
Heileman,
N.
Ghani,
and
C.
C.
Lamb,
“Qos
performance
e
v
al-
uation
of
disjoint
queue
scheduler
for
video-applications
o
v
er
lte-a
hetnets,
”
Proceedings
of
the
7th
International
Conference
on
Computing
Communication
and
Netw
orking
T
echnologies,
pp.
4,
2016.
[21]
V
.
Naumo
v
,
K.
Samouylo
v
,
N.
Y
arkina,
E.
Sopin,
S.
Andree
v
,
and
A.
Samuylo
v
,
“Lte
performance
analy-
sis
using
queuing
systems
with
finite
resources
and
random
requirements,
”
7th
International
Congress
on
Ultra
Modern
T
elecommunications
and
Control
Systems
and
W
orkshops
(ICUMT),
pp.
100–103,
2015.
[22]
R.
K.
Polag
ang
a
and
Q.
Liang,
“Self-similarity
and
modeling
of
lte/lte-a
data
traf
fic,
”
Measurement,
v
ol.
75,
pp.
218–229,
2015.
[23]
H.-P
.
Schwefel
and
L.
Lipsk
y
,
“Performance
results
for
analytic
models
of
traf
fic
in
telecommunication
systems,
based
on
multiple
ON-OFF
sources
with
self-similar
beha
vior
,
”
T
eletraf
fic
science
and
engineer
-
ing,
pp.
55–65,
1999.
Int
J
Elec
&
Comp
Eng,
V
ol.
10,
No.
5,
O
c
tober
2020
:
5420
–
5429
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Elec
&
Comp
Eng
ISSN:
2088-8708
r
5429
[24]
Ahmed
M.
Mohamed
and
,
Adel
F
.
Ag
amy
,
“
A
surv
e
y
on
the
common
netw
ork
traf
fic
sources
models,
”
Interenational
Journal
Of
Computer
Netw
ork
(IJCN),
v
ol.
3,
no.
2,
2011.
[25]
L.
Lipsk
y
,
”Queueing
Theory:
A
linear
algebraic
approach,
”
Springer
Science
Business
Media
second
Edition,
2008.
[26]
E.
Dahlman,
S.
P
arkv
all,
and
J.
Sk
old,
”4G:
L
TE/L
TE-adv
anced
for
mobile
broadband,
”
Academic
press,
2013.
BIOGRAPHIES
OF
A
UTHORS
Adel
Agamy
recei
v
ed
the
B.Sc.
in
Communications
and
Electronics
engineering
in
2007,
M.Sc.
in
Electrical
Engineering
in
2012
from
Asw
an
uni
v
ersity
,
Egypt.
He
is
currently
an
assistant
lecturer
at
electrical
engineering
department,
Asw
an
f
aculty
of
engineering,
Asw
an
Uni
v
ersity
.
His
fields
of
interest:
traf
fic
modeling,
wireless
communication,
cellular
netw
ork
and
computer
netw
ork
Ahmed
Mohamed
recei
v
ed
the
B.Sc.
in
electrical
and
c
omputer
engineering
from
Assiut
uni
v
ersity
,
Egypt,
in
1994,
the
M.Sc.
in
computer
science
and
engineering
from
uni
v
ersity
of
Connecticut,
USA
in
2001
and
the
Ph.D.
in
computer
science
and
engineering
from
uni
v
ersity
of
Connecticut,
USA
in
2004.
He
is
currently
an
assistant
professor
at
electri
cal
engineering
department,
Asw
an
f
aculty
of
engineering,
Asw
an
Uni
v
ersity
.
His
research
interests
include,
performance
modeling,
queueing
analysis,
distrib
uted
systems,
computer
netw
orks
and
operating
systems.
Influence
of
various
application
types
on
the
performance
of
...
(Adel
Agamy)
Evaluation Warning : The document was created with Spire.PDF for Python.