T
E
L
K
O
M
N
I
K
A
T
elec
o
m
m
un
ica
t
io
n Co
m
pu
t
ing
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
24
,
No
.
3
,
J
u
n
e
20
26
,
p
p
.
825
~
8
3
9
I
SS
N:
1693
-
6
9
3
0
,
DOI
: 1
0
.
1
2
9
2
8
/TE
L
KOM
NI
KA.
v
24
i
3
.
2
7
7
6
6
825
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//jo
u
r
n
a
l.u
a
d
.
a
c.
id
/in
d
ex
.
p
h
p
/TELK
OM
N
I
K
A
Neura
l ne
twork a
ppro
a
ches for q
u
a
lity
-
of
-
serv
ice op
timiza
tion
in so
ft
wa
re
-
defi
n
ed net
wo
rking
en
v
iro
nments
M
uq
a
m
ud
din
M
uh
i
b
1
,
Ra
ng
u Sridev
i
2
1
D
i
g
i
t
a
l
F
o
r
e
n
si
c
L
a
b
,
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
U
C
ESTH
,
J
a
w
a
h
a
r
l
a
l
N
e
h
r
u
Te
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
s
i
t
y
H
y
d
e
r
a
b
a
d
,
H
y
d
e
r
a
b
a
d
,
I
n
d
i
a
2
D
i
r
e
c
t
o
r
a
t
e
o
f
En
t
r
e
p
r
e
n
e
u
r
sh
i
p
,
I
n
n
o
v
a
t
i
o
n
a
n
d
S
t
a
r
t
-
u
p
s,
J
a
w
a
h
a
r
l
a
l
N
e
h
r
u
Te
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
si
t
y
H
y
d
e
r
a
b
a
d
,
H
y
d
e
r
a
b
a
d
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Dec
20
,
2
0
2
5
R
ev
is
ed
Feb
1
,
2
0
2
6
Acc
ep
ted
Mar
29
,
2
0
2
6
S
o
ftwa
re
-
d
e
fin
e
d
n
e
two
r
k
in
g
(S
DN
)
e
n
a
b
les
c
e
n
tralize
d
a
n
d
p
ro
g
ra
m
m
a
b
le
c
o
n
tro
l
o
f
n
e
two
r
k
b
e
h
a
v
i
o
r;
h
o
we
v
e
r,
c
o
n
v
e
n
ti
o
n
a
l
r
o
u
ti
n
g
stra
teg
ies
re
m
a
in
larg
e
ly
re
a
c
ti
v
e
a
n
d
stru
g
g
le
to
a
d
a
p
t
t
o
ra
p
id
ly
c
h
a
n
g
i
n
g
traffic
d
y
n
a
m
ics
.
To
a
d
d
re
ss
th
is
li
m
it
a
ti
o
n
,
th
is
stu
d
y
p
ro
p
o
se
s
a
lea
rn
in
g
-
b
a
se
d
S
DN
ro
u
ti
n
g
fra
m
e
wo
rk
t
h
a
t
i
n
t
e
g
ra
tes
a
lo
n
g
sh
o
rt
-
term
m
e
m
o
ry
(L
S
TM
)
m
o
d
e
l
t
o
p
re
d
ict
traffic
p
a
tt
e
rn
s
a
n
d
p
ro
a
c
ti
v
e
ly
o
p
ti
m
ize
ro
u
ti
n
g
d
e
c
isio
n
s.
Th
e
p
ro
p
o
se
d
a
p
p
ro
a
c
h
is
imp
lem
e
n
ted
a
n
d
e
v
a
lu
a
ted
i
n
a
n
S
D
N
tes
tb
e
d
u
sin
g
re
a
li
stic
traffic
sc
e
n
a
rio
s.
Ex
p
e
rime
n
tal
re
su
lt
s
a
re
a
v
e
ra
g
e
d
o
v
e
r
m
u
lt
ip
le
i
n
d
e
p
e
n
d
e
n
t
r
u
n
s
t
o
e
n
su
re
ro
b
u
stn
e
ss
a
n
d
re
p
r
o
d
u
c
ib
il
it
y
.
Co
m
p
a
re
d
with
sta
ti
c
sh
o
r
tes
t
-
p
a
th
ro
u
ti
n
g
a
n
d
c
las
sic
a
l
m
a
c
h
in
e
lea
rn
in
g
(M
L)
b
a
se
li
n
e
s,
t
h
e
p
r
o
p
o
se
d
m
o
d
e
l
d
e
m
o
n
stra
tes
c
o
n
siste
n
t
imp
r
o
v
e
m
e
n
ts
in
late
n
c
y
,
p
a
c
k
e
t
lo
ss
,
a
n
d
th
r
o
u
g
h
p
u
t
u
n
d
e
r
th
e
e
v
a
l
u
a
ted
c
o
n
d
it
io
n
s.
In
p
a
rti
c
u
lar,
th
e
a
b
latio
n
stu
d
y
re
p
o
rts
a
9
5
%
c
o
n
f
id
e
n
c
e
in
terv
a
l
f
o
r
e
n
d
-
to
-
e
n
d
late
n
c
y
ra
n
g
i
n
g
fro
m
5
1
.
8
t
o
5
5
.
6
m
s,
c
o
n
firmin
g
th
e
sta
ti
st
ica
l
sta
b
il
it
y
o
f
th
e
o
b
se
rv
e
d
g
a
i
n
s.
Ad
d
it
i
o
n
a
l
a
n
a
l
y
se
s
sh
o
w
t
h
a
t
th
e
f
ra
m
e
wo
rk
m
a
in
tain
s
l
o
w
i
n
fe
re
n
c
e
late
n
c
y
a
n
d
m
o
d
e
st
c
o
n
tro
l
o
v
e
rh
e
a
d
,
m
a
k
in
g
it
su
it
a
b
le
fo
r
re
a
l
-
ti
m
e
S
DN
e
n
v
ir
o
n
m
e
n
ts.
Ov
e
ra
ll
,
t
h
e
fin
d
in
g
s
in
d
ica
te
th
a
t
tem
p
o
ra
l
lea
rn
i
n
g
m
o
d
e
ls
c
a
n
e
ff
e
c
ti
v
e
ly
e
n
h
a
n
c
e
S
DN
ro
u
ti
n
g
p
e
rfo
rm
a
n
c
e
wh
e
n
e
v
a
lu
a
ted
with
i
n
c
o
n
tro
ll
e
d
e
x
p
e
rime
n
tal
se
tt
in
g
s,
o
ffe
rin
g
a
p
ra
c
ti
c
a
l
p
a
th
wa
y
t
o
wa
rd
m
o
re
a
d
a
p
ti
v
e
a
n
d
in
telli
g
e
n
t
n
e
tw
o
rk
c
o
n
tr
o
l.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
L
o
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
Netwo
r
k
o
p
tim
izatio
n
Qu
ality
o
f
s
er
v
ice
So
f
twar
e
-
d
ef
in
e
d
n
etwo
r
k
in
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Mu
q
am
u
d
d
in
Mu
h
ib
Dig
ital Fo
r
en
s
ic
L
ab
,
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scien
ce
an
d
E
n
g
in
ee
r
in
g
,
UC
E
STH
J
awa
h
ar
lal
Neh
r
u
T
ec
h
n
o
lo
g
ic
al
Un
iv
er
s
ity
Hy
d
er
ab
a
d
Ku
k
atp
ally
,
Hy
d
er
ab
ad
–
5
0
0
0
8
5
T
elan
g
an
a,
I
n
d
ia
E
m
ail: M
u
q
am
u
d
d
in
1
9
8
9
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
So
f
twar
e
-
d
ef
in
e
d
n
etwo
r
k
in
g
(
SDN)
h
as
em
er
g
ed
as
a
t
r
an
s
f
o
r
m
ativ
e
n
etwo
r
k
ar
ch
it
ec
tu
r
e
b
y
d
ec
o
u
p
lin
g
th
e
c
o
n
tr
o
l
p
lan
e
f
r
o
m
th
e
d
ata
p
lan
e
,
th
er
e
b
y
e
n
ab
lin
g
ce
n
tr
alize
d
co
n
tr
o
l,
p
r
o
g
r
am
m
a
b
ilit
y
,
an
d
f
lex
ib
le
n
etwo
r
k
m
a
n
ag
em
en
t
[
1
]
,
[
2
]
.
T
h
ese
c
h
ar
ac
ter
is
tics
m
ak
e
SDN
p
ar
ticu
lar
ly
attr
ac
tiv
e
f
o
r
m
o
d
e
r
n
telec
o
m
m
u
n
icatio
n
a
n
d
in
f
o
r
m
atio
n
an
d
c
o
m
m
u
n
icatio
n
tech
n
o
lo
g
y
(
I
C
T
)
in
f
r
astru
ct
u
r
es
th
at
d
em
an
d
ef
f
icien
t
tr
af
f
ic
en
g
i
n
ee
r
in
g
,
p
o
licy
en
f
o
r
ce
m
en
t,
a
n
d
d
y
n
am
ic
r
eso
u
r
ce
allo
ca
tio
n
[
3
]
.
As
co
n
tem
p
o
r
ar
y
n
etwo
r
k
s
in
cr
ea
s
in
g
ly
s
u
p
p
o
r
t
h
eter
o
g
e
n
eo
u
s
a
p
p
licatio
n
s
in
clu
d
in
g
in
te
r
n
et
o
f
th
in
g
s
(
I
o
T
)
s
er
v
ices,
m
u
ltime
d
ia
s
tr
ea
m
in
g
,
clo
u
d
co
m
p
u
tin
g
,
an
d
m
is
s
io
n
-
cr
iti
ca
l
s
y
s
tem
s
en
s
u
r
in
g
co
n
s
is
t
en
t
an
d
p
r
ed
ictab
le
q
u
ality
o
f
s
er
v
ice
(
Qo
S)
h
as
b
ec
o
m
e
a
f
u
n
d
am
en
tal
r
eq
u
ir
e
m
en
t
[
4
]
,
[
5
]
.
Su
ch
ap
p
licatio
n
s
im
p
o
s
e
s
tr
in
g
e
n
t
co
n
s
tr
ain
ts
o
n
laten
cy
,
th
r
o
u
g
h
p
u
t,
jitt
er
,
an
d
p
ac
k
et
lo
s
s
,
p
ar
ticu
lar
ly
u
n
d
er
d
y
n
am
ic
an
d
b
u
r
s
ty
tr
af
f
ic
co
n
d
itio
n
s
[
6
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
TEL
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
24
,
No
.
3
,
J
u
n
e
20
26
:
8
2
5
-
8
3
9
826
Desp
ite
th
e
in
h
er
en
t
f
lex
ib
ilit
y
o
f
SDN,
ef
f
ec
tiv
e
Qo
S
m
an
ag
em
en
t
r
em
ain
s
ch
allen
g
in
g
in
p
r
ac
tice.
Ma
n
y
ex
is
tin
g
SDN
-
b
ased
s
o
lu
tio
n
s
r
ely
o
n
s
tatic
r
o
u
tin
g
p
o
licies,
p
r
io
r
ity
q
u
e
u
in
g
m
ec
h
an
is
m
s
,
o
r
r
ea
ctiv
e
co
n
g
esti
o
n
co
n
t
r
o
l
s
tr
ateg
ies
[
7
]
,
[
8
]
.
T
h
ese
ap
p
r
o
ac
h
es
ty
p
ically
r
esp
o
n
d
o
n
ly
af
ter
Q
o
S
d
eg
r
ad
atio
n
h
as
alr
ea
d
y
o
cc
u
r
r
ed
,
r
esu
ltin
g
in
in
cr
ea
s
ed
laten
cy
,
p
ac
k
et
lo
s
s
,
an
d
in
ef
f
icien
t
r
eso
u
r
ce
u
tili
za
tio
n
d
u
r
in
g
tr
af
f
ic
s
u
r
g
es
[
9
]
,
[
1
0
]
.
I
n
o
p
er
atio
n
al
SDN
en
v
ir
o
n
m
e
n
ts
,
ev
en
s
h
o
r
t
-
liv
ed
co
n
g
esti
o
n
ev
en
t
s
ca
n
s
ig
n
if
ican
tly
d
eg
r
ad
e
th
e
p
er
f
o
r
m
an
ce
o
f
d
elay
-
s
en
s
itiv
e
a
n
d
b
an
d
wid
th
-
in
ten
s
iv
e
ap
p
licatio
n
s
,
u
n
d
e
r
s
co
r
in
g
th
e
lim
itatio
n
s
o
f
p
u
r
ely
r
ea
ctiv
e
Qo
S m
an
ag
em
en
t
m
ec
h
an
is
m
s
.
T
o
en
h
an
c
e
ad
ap
tab
ilit
y
,
m
ac
h
in
e
lear
n
in
g
(
ML
)
tech
n
iq
u
es
h
av
e
b
ee
n
in
cr
ea
s
in
g
ly
a
p
p
lie
d
to
SDN
-
b
ased
Qo
S
m
an
a
g
em
en
t
a
n
d
tr
af
f
ic
a
n
aly
s
is
[
1
1
]
,
[
1
2
]
.
C
las
s
ical
ML
m
o
d
els,
in
clu
d
in
g
d
ec
is
io
n
t
r
ee
s
,
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
es,
k
-
m
ea
n
s
clu
s
ter
in
g
,
an
d
r
an
d
o
m
f
o
r
ests
,
h
av
e
d
em
o
n
s
tr
ated
ef
f
ec
tiv
en
ess
in
task
s
s
u
ch
as
tr
af
f
ic
class
if
icatio
n
an
d
co
n
g
esti
o
n
d
etec
tio
n
[
1
3
]
–
[
1
5
]
.
H
o
wev
er
,
t
h
ese
m
eth
o
d
s
g
en
er
ally
ass
u
m
e
in
d
ep
en
d
en
t
o
b
s
er
v
atio
n
s
an
d
ar
e
o
f
ten
ev
alu
ate
d
th
r
o
u
g
h
o
f
f
lin
e
a
n
aly
s
is
,
wh
ich
li
m
its
th
eir
ab
ilit
y
t
o
ca
p
tu
r
e
th
e
tem
p
o
r
al
d
y
n
am
ic
s
o
f
n
etwo
r
k
t
r
af
f
ic
a
n
d
co
n
s
t
r
ain
s
th
eir
a
p
p
licab
ilit
y
in
r
ea
l
-
tim
e
SDN
co
n
tr
o
l
en
v
ir
o
n
m
en
ts
[
1
6
]
.
Mo
r
eo
v
e
r
,
th
eir
r
elian
ce
o
n
s
tatic
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
r
ed
u
ce
s
r
o
b
u
s
tn
ess
u
n
d
er
h
ig
h
ly
d
y
n
am
ic
n
etwo
r
k
c
o
n
d
itio
n
s
.
Netwo
r
k
tr
af
f
ic
in
h
er
e
n
tly
ex
h
ib
its
s
tr
o
n
g
tem
p
o
r
al
co
r
r
elatio
n
s
;
wh
er
eb
y
h
is
to
r
ical
tr
af
f
i
c
b
eh
av
io
r
d
ir
ec
tly
in
f
lu
e
n
ce
s
f
u
tu
r
e
n
et
wo
r
k
s
tates
[
1
7
]
.
I
g
n
o
r
in
g
th
e
s
e
d
ep
en
d
en
cies
r
estricts
th
e
e
f
f
ec
tiv
en
ess
o
f
Qo
S
p
r
ed
ictio
n
a
n
d
o
p
tim
izatio
n
s
tr
ateg
ies.
Dee
p
lear
n
in
g
(
DL
)
m
o
d
els,
p
a
r
ticu
lar
ly
r
ec
u
r
r
e
n
t
n
eu
r
al
n
etwo
r
k
s
(
R
NNs),
ad
d
r
ess
th
is
lim
itatio
n
b
y
lear
n
in
g
s
eq
u
en
tial
p
atter
n
s
f
r
o
m
tim
e
-
s
er
ies
d
ata
[
1
8
]
,
[
1
9
]
.
Am
o
n
g
th
ese
m
o
d
els,
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
L
STM
)
n
etwo
r
k
s
ar
e
esp
ec
ially
well
s
u
ited
f
o
r
ca
p
tu
r
in
g
l
o
n
g
-
ter
m
tem
p
o
r
al
d
ep
en
d
en
cies
wh
ile
m
itig
atin
g
th
e
v
an
is
h
in
g
g
r
ad
i
en
t
p
r
o
b
lem
[
2
0
]
.
L
STM
-
b
ased
ap
p
r
o
ac
h
es
h
av
e
s
h
o
wn
p
r
o
m
is
in
g
r
esu
lts
in
tr
af
f
ic
f
o
r
ec
asti
n
g
,
co
n
g
esti
o
n
p
r
e
d
ictio
n
,
a
n
d
a
d
ap
tiv
e
r
o
u
tin
g
[
2
1
]
–
[
2
5
]
.
Nev
er
th
eless
,
m
an
y
ex
is
tin
g
L
STM
-
d
r
iv
en
s
o
lu
tio
n
s
f
u
n
ct
io
n
as
ex
ter
n
al
o
r
o
f
f
lin
e
o
p
t
im
izatio
n
m
o
d
u
les,
r
ath
er
th
an
b
ein
g
d
ir
ec
tly
e
m
b
ed
d
e
d
with
in
th
e
SDN
c
o
n
tr
o
l
p
lan
e.
T
h
is
s
ep
a
r
atio
n
lim
its
r
ea
l
-
tim
e
r
esp
o
n
s
iv
en
ess
an
d
in
cr
ea
s
es
ar
ch
itectu
r
al
co
m
p
lex
ity
[
2
6
]
.
Ad
d
itio
n
ally
,
s
ev
er
al
DL
-
b
ased
ap
p
r
o
ac
h
es
in
tr
o
d
u
ce
n
o
n
-
tr
iv
ial
co
m
p
u
ta
tio
n
al
an
d
co
n
tr
o
ller
o
v
er
h
ea
d
,
r
aisi
n
g
co
n
ce
r
n
s
ab
o
u
t
s
ca
l
ab
ilit
y
an
d
co
n
t
r
o
l
-
p
lan
e
p
er
f
o
r
m
an
ce
in
lar
g
e
-
s
ca
le
o
r
laten
cy
-
s
en
s
itiv
e
SDN
d
ep
lo
y
m
e
n
ts
[
2
7
]
.
E
x
is
tin
g
s
tu
d
ies
also
f
r
eq
u
en
tly
r
ely
o
n
s
in
g
le
-
ap
p
licatio
n
o
r
s
y
n
th
etic
d
atasets
,
o
f
f
e
r
in
g
lim
ited
in
s
ig
h
t
in
to
s
y
s
tem
b
eh
av
io
r
u
n
d
er
r
ea
lis
tic
h
eter
o
g
en
e
o
u
s
tr
af
f
ic
co
n
d
iti
o
n
s
[
2
8
]
.
C
o
n
s
eq
u
en
tly
,
a
c
lear
r
esear
ch
g
ap
ex
is
ts
in
th
e
d
esig
n
o
f
Qo
S
o
p
tim
izatio
n
m
ec
h
a
n
is
m
s
th
at
ar
e
tem
p
o
r
ally
awa
r
e
,
tig
h
tly
in
teg
r
ate
d
in
to
th
e
S
DN
co
n
tr
o
l
p
lan
e,
co
m
p
u
tatio
n
ally
ef
f
icien
t
f
o
r
r
ea
l
-
tim
e
o
p
er
atio
n
,
an
d
v
alid
ated
u
n
d
e
r
h
eter
o
g
en
e
o
u
s
tr
af
f
ic
s
ce
n
ar
io
s
r
ep
r
esen
tativ
e
o
f
p
r
ac
tical
tele
co
m
m
u
n
icatio
n
n
etwo
r
k
s
.
T
o
a
d
d
r
e
s
s
t
h
is
g
a
p
,
t
h
is
s
t
u
d
y
p
r
o
p
o
s
e
s
a
p
r
e
d
i
ct
i
v
e
Q
o
S
o
p
t
i
m
i
z
a
ti
o
n
f
r
a
m
e
w
o
r
k
t
h
a
t
i
n
te
g
r
a
t
e
s
a
n
L
S
T
M
-
b
a
s
e
d
DL
m
o
d
e
l
d
i
r
e
c
t
l
y
w
i
t
h
i
n
t
h
e
S
D
N
c
o
n
t
r
o
l
l
e
r
.
T
h
e
p
r
o
p
o
s
e
d
f
r
a
m
e
w
o
r
k
l
e
a
r
n
s
t
e
m
p
o
r
a
l
t
r
a
f
f
i
c
p
a
t
t
e
r
n
s
f
r
o
m
s
t
r
u
c
t
u
r
e
d
f
l
o
w
-
le
v
e
l
f
e
a
t
u
r
es
a
n
d
g
e
n
e
r
a
t
es
p
r
o
a
c
t
i
v
e
r
o
u
t
i
n
g
a
n
d
r
es
o
u
r
c
e
a
llo
c
a
t
i
o
n
d
e
c
is
i
o
n
s
i
n
r
e
a
l
t
i
m
e
,
t
h
e
r
e
b
y
s
h
i
f
t
i
n
g
Q
o
S
m
a
n
a
g
e
m
e
n
t
f
r
o
m
a
r
e
a
c
t
i
v
e
t
o
a
p
r
e
d
i
c
t
i
v
e
p
a
r
a
d
i
g
m
.
B
y
a
n
t
i
c
i
p
a
t
i
n
g
f
u
t
u
r
e
n
e
t
w
o
r
k
c
o
n
d
i
t
i
o
n
s
,
t
h
e
c
o
n
tr
o
l
l
e
r
c
a
n
m
i
t
i
g
a
t
e
c
o
n
g
e
s
t
i
o
n
b
e
f
o
r
e
p
e
r
f
o
r
m
a
n
c
e
d
e
g
r
a
d
a
t
i
o
n
o
c
c
u
r
s
w
h
i
l
e
m
a
i
n
t
a
i
n
i
n
g
l
o
w
c
o
m
p
u
t
a
t
i
o
n
a
l
a
n
d
c
o
n
t
r
o
l
-
p
l
a
n
e
o
v
e
r
h
e
a
d
s
u
i
t
a
b
l
e
f
o
r
r
e
a
l
-
t
i
m
e
d
e
p
l
o
y
m
e
n
t
[
2
9
]
–
[
3
1
]
.
T
h
e
f
r
am
ewo
r
k
is
ev
alu
ated
u
s
in
g
a
co
n
tr
o
lled
SDN
test
b
ed
with
Op
en
Flo
w
-
en
ab
led
s
witch
es
an
d
m
u
ltip
le
p
u
b
lic
d
atasets
r
ep
r
e
s
en
tin
g
I
o
T
,
m
u
ltime
d
ia,
a
n
d
attac
k
tr
af
f
ic
s
ce
n
ar
io
s
.
Per
f
o
r
m
an
ce
is
ass
ess
ed
u
s
in
g
k
ey
Qo
S
m
etr
ics,
in
clu
d
in
g
en
d
-
to
-
e
n
d
laten
cy
,
th
r
o
u
g
h
p
u
t,
an
d
p
ac
k
et
lo
s
s
,
to
r
ef
lect
o
p
er
atio
n
al
r
eq
u
ir
em
e
n
ts
co
m
m
o
n
ly
e
n
co
u
n
ter
ed
in
telec
o
m
m
u
n
icatio
n
an
d
I
C
T
s
y
s
tem
s
.
T
h
e
m
ain
c
o
n
tr
ib
u
tio
n
s
o
f
th
is
s
tu
d
y
ar
e
s
u
m
m
ar
ize
d
a
s
:
(
i
)
a
p
r
ed
ictiv
e
Qo
S
o
p
tim
izatio
n
f
r
am
ewo
r
k
th
at
in
teg
r
ates
an
L
STM
m
o
d
el
d
ir
ec
tly
in
to
th
e
SDN
co
n
tr
o
l
p
lan
e
,
(
ii)
t
em
p
o
r
al
m
o
d
eli
n
g
o
f
h
eter
o
g
en
e
o
u
s
n
etwo
r
k
tr
af
f
ic
to
en
ab
le
p
r
o
ac
tiv
e
r
o
u
tin
g
an
d
r
eso
u
r
ce
allo
ca
tio
n
d
ec
is
io
n
s
,
(
iii)
c
o
m
p
r
eh
e
n
s
iv
e
ev
alu
atio
n
u
n
d
er
r
ea
lis
tic
m
ix
ed
tr
af
f
ic
s
ce
n
ar
io
s
,
in
clu
d
i
n
g
I
o
T
,
m
u
ltime
d
ia,
a
n
d
attac
k
t
r
af
f
ic
,
a
n
d
(
iv
)
d
em
o
n
s
tr
atio
n
o
f
im
p
r
o
v
e
d
Q
o
S
p
er
f
o
r
m
an
ce
with
lo
w
co
n
tr
o
ller
o
v
e
r
h
ea
d
,
s
u
p
p
o
r
tin
g
p
r
ac
tical
d
ep
lo
y
ab
ilit
y
in
SDN
-
b
ased
telec
o
m
m
u
n
icatio
n
en
v
ir
o
n
m
en
ts
.
B
y
em
b
ed
d
in
g
tem
p
o
r
al
DL
ca
p
ab
ilit
ies
in
to
SDN
-
b
ased
Qo
S
m
an
ag
em
en
t,
th
is
wo
r
k
a
d
v
an
ce
s
SDN
to
war
d
in
tellig
en
t,
p
r
ed
ictiv
e,
an
d
s
elf
-
o
p
ti
m
izin
g
n
etwo
r
k
in
g
s
y
s
tem
s
,
alig
n
in
g
with
em
e
r
g
in
g
r
eq
u
ir
em
e
n
ts
in
m
o
d
er
n
telec
o
m
m
u
n
icatio
n
,
co
m
p
u
tin
g
,
an
d
co
n
tr
o
l
-
o
r
ien
ted
n
etwo
r
k
in
f
r
astru
ctu
r
es.
2.
M
E
T
H
O
D
2
.
1
.
Sy
s
t
e
m
a
rc
hite
ct
ure
T
h
is
s
tu
d
y
p
r
o
p
o
s
es
a
clo
s
ed
-
lo
o
p
SDN
ar
ch
itectu
r
e
th
at
e
m
b
ed
s
DL
in
tellig
en
ce
d
ir
ec
tl
y
in
to
t
h
e
co
n
tr
o
l
p
lan
e
to
e
n
ab
le
p
r
ed
i
ctiv
e
Qo
S
o
p
tim
izatio
n
.
Fig
u
r
e
1
illu
s
tr
ates
th
e
o
v
e
r
all
s
y
s
tem
ar
ch
itectu
r
e,
wh
ich
co
n
s
is
ts
o
f
f
o
u
r
o
p
er
ati
o
n
al
lay
er
s
:
(
i)
t
r
af
f
ic
d
ata
ac
q
u
is
itio
n
,
(
ii)
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
p
r
ep
r
o
ce
s
s
in
g
,
(
iii)
n
eu
r
al
n
etwo
r
k
–
b
ased
Qo
S p
r
ed
ictio
n
,
a
n
d
(
i
v
)
SDN
co
n
tr
o
ller
d
ec
is
io
n
e
n
f
o
r
ce
m
en
t
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
N
eu
r
a
l n
etw
o
r
k
a
p
p
r
o
a
ch
es fo
r
q
u
a
lity
-
of
-
s
ervice
o
p
timiz
a
tio
n
in
s
o
ftw
a
r
e
-
d
efin
ed
…
(
Mu
q
a
mu
d
d
i
n
Mu
h
ib
)
827
T
r
af
f
ic
s
tatis
tics
ar
e
co
llected
f
r
o
m
th
e
d
ata
p
lan
e
t
h
r
o
u
g
h
Op
en
Flo
w
-
e
n
ab
led
s
witch
es,
wh
ich
p
er
io
d
ically
ex
p
o
r
t
f
lo
w
-
lev
el
in
f
o
r
m
atio
n
in
clu
d
in
g
p
ac
k
et
co
u
n
t,
b
y
te
co
u
n
t,
f
l
o
w
d
u
r
atio
n
,
p
r
o
to
co
l
ty
p
e,
an
d
p
o
r
t
n
u
m
b
er
s
to
th
e
SD
N
co
n
tr
o
ller
.
Data
co
llectio
n
is
p
er
f
o
r
m
e
d
u
s
in
g
th
e
c
o
n
t
r
o
ller
’
s
n
o
r
t
h
b
o
u
n
d
m
o
n
ito
r
in
g
ap
p
licatio
n
p
r
o
g
r
am
m
in
g
i
n
ter
f
ac
es
(
API
s
)
,
e
n
s
u
r
in
g
co
m
p
atib
ilit
y
with
wid
ely
u
s
ed
SDN
p
latf
o
r
m
s
s
u
ch
as
o
p
en
n
etwo
r
k
o
p
er
atin
g
s
y
s
tem
(
ONOS
)
,
R
y
u
,
an
d
Op
e
n
Day
lig
h
t.
T
h
e
ex
tr
ac
ted
f
ea
tu
r
es
ar
e
f
o
r
war
d
ed
t
o
an
em
b
ed
d
ed
n
eu
r
al
n
etwo
r
k
m
o
d
u
le
th
at
p
r
ed
icts
Qo
S
-
r
elate
d
o
u
tco
m
es.
B
ased
o
n
th
ese
p
r
ed
ictio
n
s
,
th
e
SDN
co
n
tr
o
ller
p
r
o
ac
ti
v
ely
en
f
o
r
ce
s
r
o
u
tin
g
an
d
r
eso
u
r
c
e
allo
ca
tio
n
d
ec
is
io
n
s
th
r
o
u
g
h
s
o
u
th
b
o
u
n
d
O
p
en
Flo
w
m
ess
ag
es
(
e.
g
.
,
FLOW_
MO
D)
,
co
m
p
letin
g
a
co
n
tin
u
o
u
s
f
ee
d
b
ac
k
lo
o
p
.
T
h
is
ar
c
h
itectu
r
e
s
u
p
p
o
r
ts
b
o
t
h
ce
n
tr
alize
d
an
d
lo
g
ically
d
is
tr
ib
u
ted
SDN
d
ep
lo
y
m
en
ts
wh
ile
m
ain
tain
in
g
lo
w
c
o
n
tr
o
ller
o
v
er
h
ea
d
,
e
n
s
u
r
in
g
p
r
ac
tical
d
ep
l
o
y
ab
ilit
y
.
2
.
2
.
Neura
l
net
wo
r
k
m
o
del
des
ig
n
T
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
e
m
p
lo
y
s
a
s
tack
ed
L
STM
a
r
ch
itec
tu
r
e
f
o
llo
we
d
b
y
a
f
u
lly
c
o
n
n
e
cted
(
FC
)
d
ec
is
io
n
lay
er
.
T
h
e
L
STM
la
y
er
s
m
o
d
el
tem
p
o
r
al
d
ep
en
d
en
c
ies
ac
r
o
s
s
co
n
s
ec
u
tiv
e
tr
af
f
ic
o
b
s
er
v
atio
n
s
,
wh
ile
th
e
FC
lay
er
m
ap
s
th
e
lear
n
ed
tem
p
o
r
al
r
ep
r
esen
tatio
n
s
to
Qo
S
o
u
tp
u
ts
.
T
h
e
m
o
d
el
s
u
p
p
o
r
ts
b
o
th
r
eg
r
ess
io
n
an
d
class
if
icatio
n
task
s
,
d
ep
e
n
d
in
g
o
n
th
e
o
p
tim
izatio
n
o
b
jectiv
e
;
(
i)
r
e
g
r
ess
io
n
:
p
r
e
d
ictio
n
o
f
co
n
tin
u
o
u
s
Qo
S
in
d
icato
r
s
s
u
c
h
as
ex
p
ec
t
ed
laten
cy
o
r
b
an
d
wid
th
d
e
m
an
d
,
a
n
d
(
ii)
c
lass
if
icatio
n
:
p
r
ed
ictio
n
o
f
d
is
cr
ete
r
o
u
tin
g
o
r
p
r
io
r
ity
class
es (
e.
g
.
,
lo
w
-
laten
cy
p
at
h
,
h
ig
h
-
th
r
o
u
g
h
p
u
t
p
ath
)
.
Du
r
in
g
d
e
p
lo
y
m
e
n
t,
th
e
p
r
e
d
icted
o
u
tp
u
ts
ar
e
tr
an
s
lated
i
n
to
SDN
co
n
tr
o
l
ac
tio
n
s
.
Fo
r
ex
am
p
le,
p
r
ed
icted
c
o
n
g
esti
o
n
o
r
laten
c
y
escalatio
n
tr
ig
g
e
r
s
p
r
o
ac
tiv
e
p
ath
r
ec
o
n
f
i
g
u
r
atio
n
,
wh
ile
b
an
d
wid
th
d
em
an
d
p
r
ed
ictio
n
s
g
u
id
e
q
u
e
u
e
a
n
d
r
ate
-
allo
ca
tio
n
p
o
licies.
Fig
u
r
e
1
illu
s
tr
ates
th
e
o
v
er
all
s
y
s
tem
ar
ch
itectu
r
e
o
f
th
e
L
STM
-
en
h
an
ce
d
SDN
f
r
am
e
wo
r
k
f
o
r
p
r
e
d
ictiv
e
Qo
S
o
p
tim
izatio
n
,
s
h
o
win
g
th
e
d
ata
f
lo
w
f
r
o
m
tr
af
f
i
c
co
llectio
n
,
f
ea
tu
r
e
ex
tr
ac
tio
n
,
L
STM
p
r
o
ce
s
s
in
g
,
an
d
f
in
al
S
DN
co
n
tr
o
l a
ctio
n
m
ap
p
in
g
.
Fig
u
r
e
1
.
illu
s
tr
ates th
e
o
v
e
r
all
s
y
s
tem
ar
ch
itectu
r
e
o
f
t
h
e
L
S
T
M
-
en
h
an
ce
d
SDN
f
r
am
ewo
r
k
f
o
r
p
r
ed
ictiv
e
Qo
S o
p
tim
izatio
n
2
.
3
.
T
ra
ini
ng
env
iro
nm
ent
T
h
e
n
eu
r
al
n
etwo
r
k
is
im
p
lem
en
ted
u
s
in
g
th
e
T
en
s
o
r
Flo
w/Ker
as
f
r
am
ewo
r
k
an
d
in
teg
r
ate
d
with
th
e
SDN
co
n
tr
o
ller
v
ia
R
E
STf
u
l
API
s
.
T
h
e
co
n
t
r
o
ller
p
er
io
d
ically
in
v
o
k
es
th
e
p
r
e
d
ictio
n
s
er
v
ice,
f
o
r
m
in
g
a
d
ec
is
io
n
lo
o
p
with
an
av
er
a
g
e
laten
cy
o
f
2
–
4
m
s
,
wh
ich
is
well
with
in
SD
N
co
n
tr
o
l
-
p
lan
e
tim
in
g
co
n
s
tr
ain
ts
.
T
h
e
Ad
am
o
p
tim
izer
is
u
s
ed
f
o
r
tr
ain
i
n
g
s
tab
ilit
y
.
C
r
o
s
s
-
en
tr
o
p
y
lo
s
s
is
em
p
lo
y
ed
f
o
r
class
if
icatio
n
task
s
,
wh
ile
m
ea
n
s
q
u
ar
e
d
er
r
o
r
(
MS
E
)
is
u
s
ed
f
o
r
r
eg
r
ess
io
n
task
s
.
T
a
b
l
e
1
s
u
m
m
a
r
i
z
es
t
h
e
t
r
ai
n
i
n
g
e
n
v
i
r
o
n
m
e
n
t
u
s
e
d
f
o
r
t
h
e
n
e
u
r
a
l
n
e
t
w
o
r
k
-
b
as
e
d
SD
N
Q
o
S
m
o
d
e
l
.
K
e
y
p
a
r
a
m
e
t
e
r
s
i
n
cl
u
d
e
t
h
e
T
e
n
s
o
r
F
l
o
w
/K
e
r
as
f
r
a
m
e
w
o
r
k
,
R
E
S
T
f
u
l
A
P
I
i
n
t
e
r
f
a
ce
f
o
r
S
D
N
c
o
n
t
r
o
l
l
e
r
i
n
t
e
g
r
a
ti
o
n
,
A
d
a
m
o
p
t
i
m
i
z
e
r
,
a
n
d
l
o
s
s
f
u
n
ct
i
o
n
s
(
c
r
o
s
s
-
e
n
t
r
o
p
y
f
o
r
c
la
s
s
i
f
ic
a
t
i
o
n
a
n
d
MS
E
f
o
r
r
e
g
r
e
s
s
i
o
n
)
.
T
h
e
c
o
n
f
i
g
u
r
a
t
i
o
n
e
n
s
u
r
e
s
h
i
g
h
a
c
c
u
r
a
c
y
,
l
o
w
c
o
m
p
u
t
a
t
i
o
n
a
l
o
v
e
r
h
e
a
d
,
a
n
d
a
d
ap
t
a
b
i
l
i
t
y
t
o
v
a
r
y
i
n
g
t
r
a
f
f
i
c
p
a
tt
er
n
s
.
T
ab
le
1
.
T
r
ai
n
in
g
e
n
v
ir
o
n
m
en
t
co
n
f
ig
u
r
atio
n
P
a
r
a
me
t
e
r
C
o
n
f
i
g
u
r
a
t
i
o
n
F
r
a
mew
o
r
k
Te
n
s
o
r
F
l
o
w
/
K
e
r
a
s
C
o
n
t
r
o
l
l
e
r
i
n
t
e
r
f
a
c
e
R
ESTf
u
l
A
P
I
O
p
t
i
mi
z
e
r
A
d
a
m
Lo
ss
f
u
n
c
t
i
o
n
C
r
o
ss
-
e
n
t
r
o
p
y
(
c
l
a
ssi
f
i
c
a
t
i
o
n
)
,
M
S
E
(
r
e
g
r
e
ss
i
o
n
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
TEL
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
24
,
No
.
3
,
J
u
n
e
20
26
:
8
2
5
-
8
3
9
828
2
.
4
.
Da
t
a
s
et
des
cr
iptio
n
T
o
en
s
u
r
e
r
e
p
r
o
d
u
cib
ilit
y
an
d
o
b
jectiv
e
b
en
c
h
m
ar
k
in
g
,
p
u
b
licly
av
ailab
le
SDN
an
d
n
etw
o
r
k
tr
af
f
ic
d
atasets
ar
e
u
s
ed
f
o
r
tr
ain
in
g
an
d
ev
alu
atio
n
.
Pu
b
licly
av
a
ilab
le
d
atasets
ar
e
u
s
ed
to
en
s
u
r
e
r
ep
r
o
d
u
cib
ilit
y
an
d
o
b
jectiv
e
b
en
ch
m
ar
k
in
g
.
T
h
ese
in
cl
u
d
e
UNSW
-
NB
1
5
,
C
I
C
-
I
DS2
0
1
7
,
SDN
-
s
p
ec
if
ic
s
er
v
ice
f
u
n
ctio
n
ch
ain
in
g
b
en
c
h
m
ar
k
s
,
a
n
d
T
o
N
-
I
o
T
/Ed
g
e
-
I
I
o
T
d
atasets
,
w
h
ich
co
llectiv
ely
r
ep
r
esen
t
h
e
ter
o
g
en
eo
u
s
tr
af
f
ic
s
ce
n
ar
io
s
s
u
ch
as I
o
T
,
m
u
ltime
d
ia,
n
o
r
m
al,
an
d
attac
k
tr
af
f
ic
(
T
ab
le
2
)
.
Fig
u
r
e
2
illu
s
tr
ates
th
e
p
r
o
ce
s
s
o
f
d
ataset
p
r
e
p
ar
atio
n
f
o
r
SDN
Qo
S
m
o
d
elin
g
.
R
aw
tr
af
f
i
c
f
lo
ws
ar
e
co
llected
,
clea
n
ed
,
n
o
r
m
alize
d
,
an
d
tr
an
s
f
o
r
m
e
d
in
to
s
tr
u
ctu
r
ed
f
ea
tu
r
e
v
ec
to
r
s
.
L
STM
-
co
m
p
atib
le
s
eq
u
en
ce
s
ar
e
g
en
er
ated
u
s
in
g
s
lid
in
g
w
in
d
o
ws,
an
d
Qo
S
o
u
tco
m
es
a
r
e
lab
el
-
en
co
d
e
d
.
T
h
is
p
ip
elin
e
en
s
u
r
es
tem
p
o
r
al
co
n
s
is
ten
cy
,
n
u
m
e
r
ical
s
tab
ilit
y
,
an
d
s
u
itab
ilit
y
f
o
r
n
eu
r
al
n
e
two
r
k
in
p
u
t.
T
ab
le
2
.
Data
s
et
d
escr
ip
tio
n
D
a
t
a
s
e
t
D
e
scri
p
t
i
o
n
U
N
S
W
-
N
B
1
5
M
o
d
e
r
n
a
t
t
a
c
k
+
n
o
r
m
a
l
t
r
a
f
f
i
c
C
I
C
-
I
D
S
2
0
1
7
H
i
g
h
-
q
u
a
l
i
t
y
f
l
o
w
d
a
t
a
w
i
t
h
Q
o
S
b
e
h
a
v
i
o
r
s
S
D
N
-
sp
e
c
i
f
i
c
f
l
o
w
d
a
t
a
set
(
S
F
C
b
e
n
c
h
mar
k
)
F
o
r
S
D
N
t
r
a
f
f
i
c
e
n
g
i
n
e
e
r
i
n
g
a
n
d
D
L
-
b
a
s
e
d
r
o
u
t
i
n
g
To
N
-
I
o
T
/
Ed
g
e
-
I
I
o
T
I
o
T,
N
F
V
,
a
n
d
S
D
N
t
e
l
e
me
t
r
y
s
t
r
e
a
m
s
Fig
u
r
e
2
.
illu
s
tr
ates th
e
d
ataset
p
r
ep
ar
atio
n
wo
r
k
f
lo
w
f
o
r
SDN
Qo
S m
o
d
elin
g
2
.
5
.
F
e
a
t
ure
ex
t
r
a
ct
io
n a
nd
prepro
ce
s
s
i
ng
R
aw
tr
af
f
ic
f
lo
ws
ar
e
tr
an
s
f
o
r
m
ed
in
to
s
tr
u
ctu
r
ed
f
ea
tu
r
e
v
e
cto
r
s
.
E
x
tr
ac
ted
f
ea
tu
r
es
in
clu
d
e
p
ac
k
et
co
u
n
t,
b
y
te
c
o
u
n
t,
f
lo
w
d
u
r
atio
n
,
in
ter
-
ar
r
iv
al
tim
e
,
p
r
o
to
co
l
ty
p
e,
p
o
r
t
n
u
m
b
e
r
s
,
tr
an
s
m
is
s
io
n
co
n
tr
o
l
p
r
o
to
co
l
(
T
C
P
)
win
d
o
w
s
ize,
an
d
b
an
d
wid
th
c
o
n
s
u
m
p
tio
n
.
T
h
ese
f
ea
tu
r
es
wer
e
s
elec
te
d
b
ec
au
s
e
th
ey
ar
e
(
i)
d
ir
ec
tly
av
ailab
le
f
r
o
m
SDN
f
lo
w
s
tati
s
tics
an
d
(
ii)
s
tr
o
n
g
ly
c
o
r
r
elate
d
with
Qo
S
b
eh
av
io
r
in
p
r
io
r
n
etwo
r
k
in
g
s
tu
d
ies.
Pre
p
r
o
ce
s
s
in
g
co
n
s
is
ts
o
f
th
e
f
o
llo
win
g
s
tep
s
(
T
ab
le
3
)
:
−
R
em
o
v
al
o
f
co
r
r
u
p
ted
o
r
in
co
m
p
lete
en
tr
ies
−
M
in
–
m
ax
n
o
r
m
aliza
tio
n
,
d
ef
in
ed
as
:
′
=
−
−
(
1
)
wh
er
e
is
th
e
o
r
ig
in
al
f
ea
tu
r
e
v
alu
e,
an
d
an
d
d
en
o
te
th
e
m
in
im
u
m
a
n
d
m
a
x
im
u
m
v
alu
es
o
b
s
er
v
ed
in
th
e
tr
ain
in
g
s
et
.
−
Sli
d
in
g
-
win
d
o
w
s
eq
u
e
n
ce
co
n
s
tr
u
ctio
n
f
o
r
L
STM
co
m
p
atib
i
lity
.
−
L
ab
el
en
co
d
i
n
g
o
f
Qo
S o
u
tc
o
m
es.
−
C
h
r
o
n
o
lo
g
ical
p
a
r
titi
o
n
in
g
in
t
o
tr
ain
in
g
(
7
0
%),
v
alid
atio
n
(
1
5
%),
an
d
test
in
g
(
1
5
%)
s
ets.
C
h
r
o
n
o
lo
g
ical
s
p
litt
in
g
is
ap
p
lied
to
p
r
e
v
en
t
d
ata
leak
ag
e
an
d
p
r
eser
v
e
tem
p
o
r
al
in
teg
r
ity
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
N
eu
r
a
l n
etw
o
r
k
a
p
p
r
o
a
ch
es fo
r
q
u
a
lity
-
of
-
s
ervice
o
p
timiz
a
tio
n
in
s
o
ftw
a
r
e
-
d
efin
ed
…
(
Mu
q
a
mu
d
d
i
n
Mu
h
ib
)
829
T
ab
le
3
.
Featu
r
e
e
x
tr
ac
tio
n
a
n
d
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
S
t
e
p
D
e
scri
p
t
i
o
n
D
a
t
a
c
l
e
a
n
i
n
g
R
e
m
o
v
e
c
o
r
r
u
p
t
e
d
,
m
i
ss
i
n
g
,
o
r
i
n
c
o
m
p
l
e
t
e
e
n
t
r
i
e
s
N
o
r
mal
i
z
a
t
i
o
n
A
p
p
l
y
mi
n
-
ma
x
s
c
a
l
i
n
g
:
′
=
−
−
S
e
q
u
e
n
c
e
c
o
n
st
r
u
c
t
i
o
n
U
se
a
s
l
i
d
i
n
g
-
w
i
n
d
o
w
a
p
p
r
o
a
c
h
t
o
g
e
n
e
r
a
t
e
LST
M
-
c
o
mp
a
t
i
b
l
e
se
q
u
e
n
c
e
s
La
b
e
l
e
n
c
o
d
i
n
g
En
c
o
d
e
Q
o
S
o
u
t
c
o
m
e
s a
s
i
n
t
e
g
e
r
l
a
b
e
l
s fo
r
su
p
e
r
v
i
s
e
d
l
e
a
r
n
i
n
g
D
a
t
a
s
e
t
s
p
l
i
t
P
a
r
t
i
t
i
o
n
d
a
t
a
i
n
t
o
t
r
a
i
n
i
n
g
(
7
0
%),
v
a
l
i
d
a
t
i
o
n
(
1
5
%),
a
n
d
t
e
st
i
n
g
(
1
5
%) s
e
t
s
2
.
6
.
M
o
del t
ra
ini
ng
a
nd
o
pti
m
iza
t
io
n
Mo
d
el
tr
ain
in
g
is
co
n
d
u
cted
in
a
g
r
ap
h
ics
p
r
o
ce
s
s
in
g
u
n
it
(
GPU)
-
ac
ce
ler
ated
e
n
v
ir
o
n
m
en
t.
T
o
im
p
r
o
v
e
g
e
n
er
aliza
tio
n
an
d
co
n
v
er
g
e
n
ce
s
tab
ilit
y
,
ea
r
ly
s
to
p
p
in
g
,
lear
n
in
g
-
r
ate
d
ec
ay
,
d
r
o
p
o
u
t
r
eg
u
la
r
izatio
n
,
an
d
L
2
p
e
n
alties
ar
e
ap
p
lied
.
B
atch
s
izes
ar
e
tu
n
ed
in
th
e
r
an
g
e
o
f
3
2
–
1
2
8
.
Key
h
y
p
e
r
p
a
r
am
eter
s
in
clu
d
in
g
th
e
n
u
m
b
er
o
f
L
STM
u
n
its
,
s
eq
u
en
ce
le
n
g
th
,
lear
n
in
g
r
ate,
d
r
o
p
o
u
t
r
ate,
an
d
b
atc
h
s
ize
ar
e
o
p
tim
ized
u
s
in
g
g
r
id
s
ea
r
ch
f
o
llo
wed
b
y
B
ay
esian
o
p
tim
izatio
n
.
Per
f
o
r
m
a
n
ce
r
o
b
u
s
tn
ess
is
as
s
ess
ed
ac
r
o
s
s
m
u
ltip
le
tr
af
f
ic
win
d
o
ws.
2
.
7
.
M
o
del
s
elec
t
io
n r
a
t
io
na
l
e
a
nd
im
plem
ent
a
t
io
n sa
f
eg
u
a
rds
Alth
o
u
g
h
alter
n
ativ
e
tem
p
o
r
al
m
o
d
els
s
u
ch
as
g
ated
r
ec
u
r
r
e
n
t
u
n
its
(
GR
Us
)
,
tem
p
o
r
al
co
n
v
o
lu
tio
n
al
n
etwo
r
k
s
(
T
C
Ns),
an
d
T
r
a
n
s
f
o
r
m
er
s
wer
e
co
n
s
id
er
ed
,
th
e
L
STM
ar
ch
itectu
r
e
was
s
elec
ted
b
ased
o
n
its
b
alan
ce
d
tr
a
d
e
-
o
f
f
b
etwe
en
t
em
p
o
r
al
m
o
d
elin
g
ac
cu
r
ac
y
,
in
f
er
en
ce
laten
cy
,
a
n
d
co
n
tr
o
ller
o
v
er
h
ea
d
.
Pil
o
t
co
m
p
ar
ativ
e
r
esu
lts
ar
e
p
r
o
v
id
ed
,
wh
e
r
e
L
STM
d
em
o
n
s
tr
ates
s
u
p
er
io
r
s
u
itab
ilit
y
f
o
r
SDN
co
n
tr
o
l
-
p
la
n
e
d
ep
lo
y
m
e
n
t.
T
r
an
s
f
o
r
m
er
-
b
a
s
ed
m
o
d
els
ex
h
ib
ited
h
ig
h
e
r
in
f
er
en
ce
laten
cy
,
wh
ic
h
v
io
lates
r
ea
l
-
tim
e
ap
p
licab
ilit
y
co
n
s
tr
ain
ts
,
wh
ile
GR
Us
s
h
o
wed
s
lig
h
tly
lo
we
r
ac
cu
r
ac
y
u
n
d
er
h
eter
o
g
en
e
o
u
s
tr
af
f
ic
c
o
n
d
itio
n
s
.
T
o
p
r
ev
en
t
d
ata
leak
a
g
e,
all
d
atasets
ar
e
p
ar
titi
o
n
ed
ch
r
o
n
o
l
o
g
ically
,
an
d
s
lid
in
g
win
d
o
ws
ar
e
ap
p
lied
s
tr
ictly
with
in
ea
ch
p
ar
titi
o
n
.
T
im
estam
p
au
d
its
co
n
f
ir
m
th
e
a
b
s
en
ce
o
f
lo
o
k
-
ah
ea
d
b
ias.
2
.
8
.
Ra
t
io
na
le
f
o
r
s
elec
t
ing
L
S
T
M
T
o
s
u
b
s
tan
tiate
th
is
ch
o
ice,
we
co
n
d
u
cted
p
ilo
t
ex
p
er
im
e
n
ts
o
n
a
m
ix
e
d
-
tr
af
f
ic
s
u
b
s
et
o
f
C
I
C
-
I
DS2
0
1
7
an
d
T
o
N
-
I
o
T
,
co
m
p
r
is
in
g
1
0
,
0
0
0
tem
p
o
r
al
s
eq
u
en
ce
s
.
T
ab
le
4
s
u
m
m
ar
izes
th
e
co
m
p
ar
ativ
e
r
esu
lts
.
T
ab
le
4
p
r
esen
ts
a
co
m
p
ar
ativ
e
an
aly
s
is
o
f
s
ev
er
al
tem
p
o
r
al
ar
ch
itectu
r
es
ev
alu
ated
o
v
e
r
f
iv
e
ex
p
er
im
en
tal
r
u
n
s
,
h
ig
h
lig
h
tin
g
k
ey
p
er
f
o
r
m
an
ce
m
etr
ics
s
u
ch
as
ac
cu
r
ac
y
,
1
-
s
co
r
e,
in
f
er
e
n
ce
laten
c
y
,
p
ar
am
eter
co
u
n
t,
an
d
co
n
tr
o
ller
o
v
er
h
ea
d
.
T
h
e
r
esu
lts
in
d
icate
th
at
wh
ile
th
e
GR
U
m
o
d
el
o
f
f
er
s
s
lig
h
tl
y
lo
wer
i
n
f
er
en
ce
laten
cy
(
1
.
9
m
s
)
an
d
r
ed
u
ce
d
p
ar
am
eter
c
o
u
n
t
(
0
.
3
3
M)
,
i
t
s
u
f
f
er
s
a
1
.
2
%
d
r
o
p
in
ac
c
u
r
ac
y
co
m
p
ar
e
d
to
L
STM
,
wh
ich
m
ay
b
e
cr
itical
f
o
r
Qo
S
-
s
en
s
itiv
e
s
ce
n
ar
io
s
wh
er
e
m
is
class
if
icatio
n
s
ca
n
tr
ig
g
er
u
n
n
ec
ess
ar
y
r
er
o
u
tin
g
o
r
s
er
v
ice
d
e
g
r
ad
at
io
n
.
T
C
N
m
o
d
els,
d
esp
ite
m
ain
tain
in
g
s
tab
le
p
er
f
o
r
m
an
c
e
u
n
d
er
s
tatio
n
ar
y
tr
af
f
ic
co
n
d
itio
n
s
,
d
em
o
n
s
tr
at
e
lim
ited
r
esp
o
n
s
iv
en
ess
to
b
u
r
s
ty
I
o
T
wo
r
k
l
o
ad
s
d
u
e
to
t
h
eir
f
ix
ed
r
ec
e
p
tiv
e
f
ield
d
esig
n
.
T
h
e
T
r
an
s
f
o
r
m
e
r
ar
ch
itectu
r
e
,
alth
o
u
g
h
co
m
p
etitiv
e
in
ac
cu
r
ac
y
(
9
2
.
6
%),
i
n
cu
r
s
s
u
b
s
tan
tially
h
ig
h
er
in
f
er
en
ce
laten
c
y
(
1
4
.
8
m
s
)
an
d
c
o
n
tr
o
ller
o
v
er
h
ea
d
(
1
8
.
2
%),
ex
ce
e
d
in
g
ty
p
ical
SDN
co
n
tr
o
l
-
p
lan
e
tim
in
g
co
n
s
tr
ain
ts
o
f
5
–
1
0
m
s
an
d
r
aisi
n
g
s
ca
lab
ilit
y
co
n
ce
r
n
s
lin
k
ed
to
its
q
u
ad
r
atic
atten
tio
n
m
ec
h
a
n
is
m
.
I
n
co
n
tr
ast,
th
e
L
STM
(
2
×1
2
8
+FC
)
m
o
d
el
p
r
o
v
id
es
th
e
m
o
s
t
b
alan
ce
d
p
er
f
o
r
m
an
ce
,
a
ch
iev
in
g
n
ea
r
-
p
ea
k
ac
cu
r
ac
y
(
9
4
.
3
%)
with
s
u
b
-
3
m
s
laten
cy
an
d
co
n
tr
o
ller
o
v
er
h
ea
d
b
elo
w
9
%,
alig
n
in
g
cl
o
s
ely
with
o
p
er
atio
n
al
r
eq
u
ir
em
e
n
ts
in
SDN
en
v
ir
o
n
m
en
ts
as r
ep
o
r
ted
i
n
co
n
tem
p
o
r
ar
y
s
tu
d
ies.
T
ab
le
4
.
C
o
m
p
a
r
is
o
n
o
f
tem
p
o
r
al
ar
ch
itectu
r
e
(
m
ea
n
in
g
5
r
u
n
s
)
M
o
d
e
l
A
c
c
u
r
a
c
y
(
%)
1
-
sc
o
r
e
I
n
f
e
r
e
n
c
e
l
a
t
e
n
c
y
(
ms)
P
a
r
a
me
t
e
r
s (
M
)
C
o
n
t
r
o
l
l
e
r
o
v
e
r
h
e
a
d
(
%)
LSTM
(
2
×
1
2
8
F
C
)
9
4
.
3
3
.
9
2
.
1
0
.
4
1
8
.
3
G
R
U
(
2
×
1
2
8
+
F
C
)
9
3
.
1
2
.
7
1
.
9
0
.
3
3
7
.
8
TC
N
(
3
l
a
y
e
r
s,
k
=
7
)
9
1
.
8
0
.
5
3
.
4
0
.
5
2
9
.
7
Tr
a
n
sf
o
r
mer (
2
-
l
a
y
e
r
)
9
2
.
6
2
.
0
1
4
.
8
1
.
7
3
1
8
.
2
2
.
9
.
Sa
f
eg
ua
rds
a
g
a
ins
t
da
t
a
lea
k
a
g
e
a
nd
a
da
pta
bil
it
y
T
o
p
r
ev
en
t d
ata
leak
a
g
e
an
d
e
n
s
u
r
e
m
eth
o
d
o
lo
g
ical
r
ig
o
r
,
al
l d
atasets
ar
e
p
ar
titi
o
n
ed
ch
r
o
n
o
lo
g
ically
r
ath
er
th
a
n
r
a
n
d
o
m
l
y
.
T
r
ai
n
in
g
,
v
alid
atio
n
,
an
d
test
in
g
s
ets
ar
e
d
er
iv
e
d
f
r
o
m
n
o
n
-
o
v
e
r
lap
p
in
g
tim
e
win
d
o
ws,
an
d
s
lid
in
g
win
d
o
ws ar
e
ap
p
li
ed
s
tr
ictly
with
in
ea
ch
p
ar
titi
o
n
.
T
im
estam
p
au
d
its
co
n
f
ir
m
t
h
e
ab
s
en
ce
o
f
lo
o
k
-
ah
ea
d
b
ias.
T
o
m
ain
tain
r
o
b
u
s
tn
ess
u
n
d
er
ev
o
lv
in
g
tr
af
f
ic
co
n
d
itio
n
s
,
an
ex
p
o
n
en
ti
al
m
o
v
in
g
av
er
a
g
e
(
E
MA
)
–
b
ased
o
n
lin
e
n
o
r
m
ali
za
tio
n
s
ch
em
e
is
em
p
lo
y
ed
d
u
r
in
g
d
e
p
lo
y
m
e
n
t.
T
h
is
ap
p
r
o
ac
h
en
ab
les
g
r
ad
u
al
ad
ap
tatio
n
to
tr
af
f
ic
d
is
tr
ib
u
ti
o
n
s
h
if
ts
wh
ile
lim
itin
g
d
r
if
t.
Stre
s
s
test
in
g
u
s
in
g
in
jecte
d
o
u
tlier
tr
af
f
ic
s
h
o
ws
m
in
im
al
p
er
f
o
r
m
an
ce
d
eg
r
a
d
a
tio
n
co
m
p
a
r
ed
to
s
tatic
n
o
r
m
al
izatio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
TEL
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
24
,
No
.
3
,
J
u
n
e
20
26
:
8
2
5
-
8
3
9
830
2
.
1
0
.
O
nli
ne
no
rm
a
liza
t
io
n
a
nd
a
da
pta
bil
it
y
T
o
m
ain
tain
r
o
b
u
s
tn
ess
u
n
d
e
r
ev
o
lv
i
n
g
tr
a
f
f
ic
d
is
tr
ib
u
tio
n
s
,
an
E
MA
–
b
ased
o
n
lin
e
n
o
r
m
aliza
tio
n
s
ch
em
e
is
ap
p
lied
d
u
r
in
g
d
ep
lo
y
m
en
t.
Un
lik
e
s
tatic
m
in
–
m
ax
s
ca
lin
g
,
th
is
ap
p
r
o
ac
h
en
ab
les
g
r
ad
u
al
ad
ap
tatio
n
to
tr
af
f
ic
s
h
if
ts
wh
ile
lim
itin
g
d
is
tr
ib
u
tio
n
d
r
if
t.
S
tr
ess
test
in
g
with
in
jecte
d
o
u
tlier
tr
af
f
ic
r
esu
lted
in
m
in
im
al
p
er
f
o
r
m
a
n
ce
d
e
g
r
a
d
atio
n
,
d
e
m
o
n
s
tr
atin
g
a
d
ap
tab
ilit
y
u
n
d
er
r
ea
lis
tic
co
n
d
itio
n
s
.
2
.
1
1
.
E
v
a
lua
t
io
n
m
e
t
rics
Mo
d
el
p
er
f
o
r
m
an
ce
is
ev
alu
a
ted
at
b
o
th
p
r
ed
ictio
n
a
n
d
n
e
two
r
k
lev
els.
Pre
d
ictio
n
p
er
f
o
r
m
an
ce
is
ass
es
s
ed
u
s
in
g
class
if
icatio
n
m
etr
ics
(
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
r
ec
ei
v
er
o
p
e
r
atin
g
ch
ar
ac
te
r
is
tic
(
R
OC
)
-
ar
ea
u
n
d
er
th
e
cu
r
v
e
(
AUC
)
)
an
d
r
eg
r
ess
io
n
m
etr
i
cs
(
MSE
,
m
ea
n
ab
s
o
lu
te
er
r
o
r
(
MA
E
)
,
an
d
R²
)
.
Netwo
r
k
-
lev
el
p
e
r
f
o
r
m
an
ce
i
s
ev
alu
ated
u
s
in
g
en
d
-
to
-
e
n
d
laten
cy
im
p
r
o
v
em
en
t,
t
h
r
o
u
g
h
p
u
t
en
h
a
n
ce
m
en
t,
p
ac
k
et
lo
s
s
r
atio
,
p
ath
s
etu
p
tim
e,
an
d
c
o
n
tr
o
ller
o
v
er
h
ea
d
(
T
ab
le
5
)
.
T
o
g
eth
er
,
t
h
ese
m
etr
ics
p
r
o
v
id
e
a
co
m
p
r
eh
en
s
iv
e
ass
ess
m
en
t
o
f
Qo
S
ef
f
ec
tiv
en
ess
,
co
n
tr
o
ller
ef
f
icien
cy
,
an
d
s
y
s
tem
-
lev
el
im
p
ac
t.
T
ab
le
5
p
r
esen
ts
th
e
m
etr
ics
u
s
ed
to
ev
alu
ate
m
o
d
el
p
er
f
o
r
m
an
ce
at
b
o
th
Qo
S
p
r
ed
ictio
n
an
d
n
etwo
r
k
p
er
f
o
r
m
an
ce
lev
els.
C
lass
if
icatio
n
m
etr
ics
in
clu
d
e
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
co
n
f
u
s
io
n
m
atr
ix
,
a
n
d
R
OC
-
AU
C
.
R
eg
r
es
s
io
n
m
etr
ics
in
clu
d
e
MSE
,
MA
E
,
an
d
R
².
Netwo
r
k
-
lev
el
m
etr
ics ass
es
s
laten
cy
,
th
r
o
u
g
h
p
u
t,
p
ac
k
et
lo
s
s
,
p
ath
s
etu
p
ti
m
e,
an
d
c
o
n
tr
o
l
o
v
er
h
ea
d
.
T
ab
le
5
.
E
v
alu
atio
n
m
et
r
ics
M
e
t
r
i
c
t
y
p
e
M
e
t
r
i
c
P
u
r
p
o
se
C
l
a
s
si
f
i
c
a
t
i
o
n
A
c
c
u
r
a
c
y
O
v
e
r
a
l
l
c
o
r
r
e
c
t
n
e
s
s
P
r
e
c
i
s
i
o
n
P
r
e
d
i
c
t
i
o
n
q
u
a
l
i
t
y
R
e
c
a
l
l
S
e
n
s
i
t
i
v
i
t
y
F1
-
sc
o
r
e
B
a
l
a
n
c
e
d
m
e
t
r
i
c
C
o
n
f
u
s
i
o
n
m
a
t
r
i
x
C
l
a
s
s
d
i
s
t
r
i
b
u
t
i
o
n
R
O
C
-
AUC
D
i
scri
m
i
n
a
t
i
o
n
a
b
i
l
i
t
y
R
e
g
r
e
ssi
o
n
M
S
E
P
e
n
a
l
i
z
e
s
l
a
r
g
e
e
r
r
o
r
s
M
A
E
M
e
a
su
r
e
s a
b
s
o
l
u
t
e
d
e
v
i
a
t
i
o
n
s
R
²
s
c
o
r
e
V
a
r
i
a
n
c
e
e
x
p
l
a
n
a
t
i
o
n
N
e
t
w
o
r
k
-
l
e
v
e
l
En
d
-
to
-
e
n
d
l
a
t
e
n
c
y
i
m
p
r
o
v
e
m
e
n
t
(
%)
Q
o
S
e
n
h
a
n
c
e
me
n
t
Th
r
o
u
g
h
p
u
t
e
n
h
a
n
c
e
m
e
n
t
(
%)
Ef
f
i
c
i
e
n
c
y
g
a
i
n
P
a
c
k
e
t
l
o
ss ra
t
i
o
R
e
l
i
a
b
i
l
i
t
y
P
a
t
h
se
t
u
p
t
i
m
e
R
o
u
t
i
n
g
e
f
f
i
c
i
e
n
c
y
C
o
n
t
r
o
l
l
e
r
o
v
e
r
h
e
a
d
S
D
N
p
e
r
f
o
r
m
a
n
c
e
i
m
p
a
c
t
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
SDN
t
estbed
c
o
nfig
ura
t
i
o
n
T
h
e
ex
p
er
im
e
n
tal
s
etu
p
was
i
m
p
lem
en
ted
o
n
a
co
n
tr
o
lled
S
DN
test
b
ed
d
esig
n
ed
to
s
im
u
l
ate
r
ea
lis
tic
n
etwo
r
k
en
v
i
r
o
n
m
e
n
ts
f
o
r
ev
alu
atin
g
Qo
S
o
p
tim
izatio
n
te
ch
n
iq
u
es.
T
h
e
test
b
ed
co
n
s
is
ted
o
f
Op
en
Flo
w
-
en
ab
led
s
witch
es c
o
n
n
ec
ted
to
a
ce
n
tr
al
SDN
co
n
tr
o
ller
,
im
p
lem
en
ted
o
n
a
h
ig
h
-
p
er
f
o
r
m
an
ce
s
er
v
er
eq
u
ip
p
ed
with
an
NVI
DI
A
GPU
to
ac
ce
ler
ate
L
STM
m
o
d
el
tr
ain
in
g
.
Ho
s
ts
in
th
e
n
etwo
r
k
g
en
er
ated
m
ix
ed
tr
af
f
i
c
p
atter
n
s
,
in
clu
d
i
n
g
h
y
p
er
te
x
t
tr
an
s
f
er
p
r
o
to
c
o
l
(
HT
T
P
)
,
v
o
ice
o
v
er
in
ter
n
et
p
r
o
to
co
l
(
Vo
I
P
)
,
a
n
d
v
id
e
o
s
tr
ea
m
in
g
f
lo
ws,
to
em
u
late
r
e
al
-
wo
r
ld
n
etwo
r
k
co
n
d
itio
n
s
.
Key
co
m
p
o
n
en
ts
o
f
t
h
e
test
b
ed
in
clu
d
e
d
:
−
SDN
C
o
n
tr
o
ller
: O
NOS
an
d
R
y
u
p
latf
o
r
m
s
wer
e
u
s
ed
f
o
r
r
u
l
e
en
f
o
r
ce
m
en
t a
n
d
f
lo
w
m
o
n
it
o
r
in
g
.
−
Switch
es: Op
en
v
Switch
(
OVS)
in
s
tan
ce
s
im
p
lem
en
ted
o
n
v
ir
tu
alize
d
m
ac
h
in
es.
−
T
r
af
f
ic
Gen
er
atio
n
:
iPer
f
a
n
d
cu
s
to
m
Py
th
o
n
s
cr
ip
ts
g
en
er
ated
v
ar
iab
le
tr
af
f
ic
lo
a
d
s
to
s
im
u
late
b
o
th
n
o
r
m
al
an
d
co
n
g
ested
n
etwo
r
k
s
ce
n
ar
io
s
.
−
Mo
n
ito
r
in
g
T
o
o
ls
: Wi
r
esh
ar
k
an
d
th
e
co
n
tr
o
ller
’
s
No
r
th
b
o
u
n
d
API
co
llected
p
ac
k
et
-
lev
el
s
tatis
t
ics,
s
u
ch
as f
lo
w
d
u
r
atio
n
,
in
ter
ar
r
iv
al
tim
es,
b
an
d
wid
th
u
s
ag
e,
an
d
p
a
ck
et
lo
s
s
.
T
h
e
test
b
ed
s
u
p
p
o
r
ted
d
y
n
am
ic
ev
alu
atio
n
o
f
m
o
d
el
p
r
ed
ic
tio
n
s
,
en
ab
lin
g
r
ea
l
-
tim
e
co
m
p
ar
is
o
n
o
f
s
tatic
r
o
u
tin
g
,
s
u
p
p
o
r
t
v
ec
t
o
r
m
ac
h
in
es
(
SVM
)
-
b
ased
r
o
u
t
in
g
,
an
d
L
STM
-
b
ased
Qo
S
o
p
tim
izatio
n
.
T
h
is
d
esig
n
en
s
u
r
ed
r
ep
r
o
d
u
cib
ilit
y
an
d
s
ca
lab
ilit
y
f
o
r
v
ar
io
u
s
tr
af
f
ic
lo
ad
s
an
d
n
etwo
r
k
t
o
p
o
lo
g
ies.
Fig
u
r
e
3
illu
s
tr
ates
th
e
en
d
-
to
-
en
d
wo
r
k
f
lo
w
o
f
th
e
p
r
o
p
o
s
ed
L
STM
-
b
ased
Qo
S
p
r
e
d
ictio
n
a
n
d
S
DN
co
n
tr
o
l
s
y
s
tem
.
Netwo
r
k
tr
af
f
ic
d
ata
is
c
o
llected
f
r
o
m
th
e
d
ata
p
lan
e,
e
n
co
m
p
ass
in
g
f
ea
tu
r
es
s
u
ch
as
p
ac
k
et
co
u
n
t,
f
lo
w
d
u
r
atio
n
,
in
ter
ar
r
i
v
al
tim
es,
T
C
P
win
d
o
w
s
ize,
an
d
p
r
o
t
o
co
l
in
f
o
r
m
atio
n
.
T
h
e
f
ea
tu
r
e
ex
tr
ac
to
r
m
o
d
u
le
tr
an
s
f
o
r
m
s
r
aw
tr
a
f
f
ic
in
to
s
tr
u
ctu
r
ed
v
ec
to
r
s
s
u
itab
le
f
o
r
n
e
u
r
al
n
etwo
r
k
in
p
u
t.
Pre
p
r
o
ce
s
s
ed
f
ea
tu
r
es
ar
e
f
ed
in
to
an
L
STM
n
etwo
r
k
,
wh
ic
h
ca
p
tu
r
es
tem
p
o
r
al
p
atter
n
s
an
d
d
e
p
en
d
e
n
cies
in
tr
a
f
f
ic
f
l
o
ws
to
p
r
e
d
ict
Qo
S
p
ar
am
eter
s
s
u
ch
as late
n
cy
,
t
h
r
o
u
g
h
p
u
t,
o
r
o
p
tim
al
r
o
u
tin
g
p
ath
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
N
eu
r
a
l n
etw
o
r
k
a
p
p
r
o
a
ch
es fo
r
q
u
a
lity
-
of
-
s
ervice
o
p
timiz
a
tio
n
in
s
o
ftw
a
r
e
-
d
efin
ed
…
(
Mu
q
a
mu
d
d
i
n
Mu
h
ib
)
831
T
h
e
SDN
co
n
t
r
o
ller
r
ec
eiv
es
th
ese
p
r
e
d
ictio
n
s
v
ia
th
e
s
o
u
th
b
o
u
n
d
API
an
d
d
y
n
am
ical
ly
u
p
d
ates
f
lo
w
r
u
les
o
n
n
etwo
r
k
d
ev
ice
s
,
cr
ea
tin
g
a
clo
s
ed
-
lo
o
p
s
y
s
tem
th
at
co
n
tin
u
o
u
s
ly
ad
ap
ts
to
ch
an
g
in
g
n
etwo
r
k
co
n
d
itio
n
s
.
T
h
is
in
teg
r
atio
n
h
ig
h
lig
h
ts
th
e
p
r
ed
ictiv
e
an
d
a
d
ap
tiv
e
ca
p
ab
ilit
ies
o
f
t
h
e
f
r
a
m
ewo
r
k
,
en
s
u
r
in
g
ef
f
icien
t tr
af
f
ic
r
o
u
tin
g
an
d
r
eso
u
r
ce
allo
ca
tio
n
in
r
ea
l
-
tim
e.
Fig
u
r
e
3
.
L
STM
-
b
ased
Qo
S
p
r
ed
ictio
n
wo
r
k
f
lo
w
3
.
2
.
SDN
t
estbed
co
nfig
ura
t
i
o
n
T
h
e
ex
p
er
im
en
ts
wer
e
c
o
n
d
u
cted
o
n
a
co
n
tr
o
lled
SDN
test
b
ed
co
n
s
is
tin
g
o
f
Op
en
Flo
w
-
en
ab
led
s
witch
es
an
d
ONO
S/
R
y
u
co
n
tr
o
ller
s
r
u
n
n
in
g
o
n
a
GPU
-
eq
u
ip
p
e
d
s
er
v
er
.
Ho
s
ts
g
en
er
ated
h
eter
o
g
en
e
o
u
s
tr
af
f
ic,
in
clu
d
i
n
g
HT
T
P,
Vo
I
P,
m
u
ltime
d
ia
s
tr
ea
m
in
g
,
a
n
d
I
o
T
f
lo
ws.
W
ir
esh
ar
k
a
n
d
c
o
n
tr
o
ller
API
s
wer
e
u
s
ed
to
co
llect
f
lo
w
s
tatis
tics
.
T
h
is
s
etu
p
s
u
p
p
o
r
ts
r
ea
l
-
tim
e
ev
alu
atio
n
o
f
s
tatic
r
o
u
tin
g
,
SVM
-
b
ased
ad
ap
tiv
e
r
o
u
tin
g
,
an
d
L
STM
-
b
ased
p
r
e
d
ictiv
e
r
o
u
tin
g
,
en
a
b
lin
g
r
e
p
r
o
d
u
cib
ilit
y
an
d
s
ca
lab
ilit
y
ac
r
o
s
s
tr
af
f
ic
s
ce
n
ar
io
s
.
3
.
3
.
P
er
f
o
r
m
a
nce
m
e
t
rics e
v
a
lua
t
io
n
T
o
ass
ess
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
L
STM
-
b
ased
Qo
S
p
r
e
d
ictio
n
,
s
ev
er
al
p
er
f
o
r
m
a
n
ce
m
et
r
ics
wer
e
ev
alu
ated
at
b
o
th
th
e
m
o
d
el
an
d
n
etwo
r
k
lev
els.
T
h
ese
m
etr
ics
p
r
o
v
id
ed
in
s
ig
h
ts
in
to
p
r
ed
ictiv
e
ac
cu
r
ac
y
,
tr
af
f
ic
m
an
ag
e
m
en
t e
f
f
icien
cy
,
an
d
o
v
er
all
n
etwo
r
k
p
e
r
f
o
r
m
an
ce
.
T
ab
le
6
p
r
o
v
id
es
an
o
v
er
all
c
o
m
p
ar
is
o
n
ac
r
o
s
s
all
tr
af
f
ic
f
lo
ws,
h
ig
h
lig
h
tin
g
h
o
w
th
e
L
S
T
M
-
b
ased
p
r
ed
ictiv
e
r
o
u
tin
g
o
u
tp
er
f
o
r
m
s
s
tatic
an
d
SV
M
-
b
ased
ap
p
r
o
ac
h
es.
C
lass
if
icat
io
n
m
etr
ics
s
h
o
w
th
at
L
STM
ac
h
iev
es 9
4
.
3
% a
cc
u
r
ac
y
an
d
0
.
9
3
9
F1
-
s
co
r
e,
d
em
o
n
s
tr
atin
g
its
s
tr
o
n
g
tr
af
f
ic
s
tate
d
is
cr
im
in
atio
n
.
R
eg
r
ess
io
n
m
etr
ics
(
MSE
,
MA
E
,
R
²)
co
n
f
ir
m
L
STM
’
s
s
u
p
er
io
r
ab
ilit
y
to
p
r
ed
ict
Qo
S
p
ar
am
eter
s
ac
cu
r
ately
.
At
th
e
n
etwo
r
k
lev
el,
laten
cy
is
r
e
d
u
ce
d
b
y
m
o
r
e
th
an
5
0
%,
th
r
o
u
g
h
p
u
t
i
m
p
r
o
v
es
b
y
o
v
e
r
5
0
%,
an
d
p
ac
k
et
lo
s
s
is
m
in
im
ized
,
all
wh
ile
k
ee
p
in
g
co
n
tr
o
ller
o
v
er
h
ea
d
lo
w
(
8
.
3
%
)
.
T
h
is
tab
le
s
u
m
m
ar
izes a
g
g
r
eg
ated
p
er
f
o
r
m
an
ce
f
o
r
th
e
e
n
tire
n
etwo
r
k
,
s
er
v
in
g
as th
e
b
aselin
e
ev
alu
atio
n
.
T
ab
le
6
.
C
o
m
p
r
eh
en
s
iv
e
p
er
f
o
r
m
an
ce
co
m
p
ar
is
o
n
o
f
s
tatic
r
o
u
tin
g
,
SVM
-
b
ased
a
d
ap
tiv
e
r
o
u
tin
g
,
a
n
d
L
STM
-
b
ased
p
r
ed
ictiv
e
r
o
u
tin
g
M
e
t
r
i
c
c
a
t
e
g
o
r
y
M
e
t
r
i
c
S
t
a
t
i
c
r
o
u
t
i
n
g
S
V
M
LSTM
(
o
u
r
s)
I
mp
r
o
v
e
m
e
n
t
v
s.
s
t
a
t
i
c
(
%)
C
l
a
s
si
f
i
c
a
t
i
o
n
A
c
c
u
r
a
c
y
(
%)
7
8
.
2
8
5
.
6
9
4
.
3
+
1
6
.
1
P
r
e
c
i
s
i
o
n
(
%)
7
6
.
5
8
4
.
1
9
3
.
7
+
1
7
.
2
R
e
c
a
l
l
(
%)
7
7
.
0
8
5
.
0
9
4
.
1
+
1
7
.
1
F1
-
sc
o
r
e
0
.
7
6
7
0
.
8
4
5
0
.
9
3
9
+
2
2
.
4
R
O
C
-
AUC
0
.
8
1
0
.
8
9
0
.
9
6
+
1
8
.
5
R
e
g
r
e
ssi
o
n
M
S
E
0
.
0
1
2
0
.
0
0
4
−
6
6
.
7
M
A
E
0
.
0
8
9
0
.
0
3
2
−
6
4
.
0
R²
0
.
8
7
0
.
9
5
+
9
.
2
N
e
t
w
o
r
k
-
l
e
v
e
l
La
t
e
n
c
y
(
ms)
1
4
5
.
7
1
0
2
.
3
6
7
.
4
−
5
3
.
7
Th
r
o
u
g
h
p
u
t
(
M
b
p
s)
8
5
0
1
0
2
5
1
2
8
0
+
5
0
.
6
P
a
c
k
e
t
l
o
ss (%)
3
.
4
2
.
1
0
.
8
−
7
6
.
5
P
a
t
h
se
t
u
p
t
i
m
e
(
ms)
1
8
.
7
1
4
.
5
9
.
2
−
5
0
.
8
C
o
n
t
r
o
l
l
e
r
o
v
e
r
h
e
a
d
(
%)
1
2
.
5
1
0
.
1
8
.
3
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
TEL
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
24
,
No
.
3
,
J
u
n
e
20
26
:
8
2
5
-
8
3
9
832
Fig
u
r
e
4
v
is
u
alize
s
r
eg
r
ess
io
n
m
etr
ics
(
MSE
,
MA
E
,
R
2
)
,
h
i
g
h
lig
h
tin
g
L
STM
’
s
s
u
p
er
io
r
p
r
ed
ictiv
e
ac
cu
r
ac
y
co
m
p
a
r
ed
to
SVM.
Hig
h
R
²
v
alu
es
in
d
icate
t
h
at
L
STM
ef
f
ec
tiv
ely
ca
p
tu
r
es
tr
a
f
f
ic
v
a
r
ian
ce
,
w
h
ich
is
cr
u
cial
f
o
r
o
p
tim
izin
g
laten
cy
an
d
th
r
o
u
g
h
p
u
t.
T
h
e
L
ST
M
m
o
d
el
ac
h
iev
e
d
s
ig
n
if
ican
t
ly
lo
wer
MSE
an
d
MA
E
co
m
p
ar
ed
to
SVM,
in
d
icatin
g
m
o
r
e
p
r
ec
is
e
Qo
S
p
r
ed
ictio
n
.
Hig
h
R
²
v
alu
es
co
n
f
ir
m
th
at
L
STM
ca
p
tu
r
es
v
ar
ia
n
ce
in
n
etwo
r
k
tr
af
f
ic
b
etter
th
an
s
tatic
o
r
S
VM
ap
p
r
o
ac
h
es,
wh
ich
is
cr
itical
f
o
r
laten
c
y
a
n
d
th
r
o
u
g
h
p
u
t
o
p
tim
izatio
n
.
Fig
u
r
e
4
.
R
eg
r
ess
io
n
m
etr
ics
f
o
r
Qo
S
p
r
e
d
ictio
n
3
.
4
.
Co
m
pa
ra
t
iv
e
a
na
ly
s
is
:
s
t
a
t
ic
v
s
.
SVM
v
s
.
L
ST
M
Fig
u
r
e
5
p
r
esen
ts
a
c
o
m
p
ar
ati
v
e
ev
alu
atio
n
o
f
s
tatic
r
o
u
tin
g
,
SVM
-
b
ased
r
o
u
tin
g
,
an
d
L
S
T
M
-
b
ased
p
r
ed
ictiv
e
r
o
u
tin
g
in
th
e
SDN
en
v
ir
o
n
m
en
t.
Static
r
o
u
tin
g
r
elies
o
n
p
r
e
d
ef
in
ed
p
ath
s
,
wh
i
ch
d
o
n
o
t
ad
ap
t
t
o
ch
an
g
in
g
tr
af
f
ic
co
n
d
itio
n
s
,
r
esu
ltin
g
in
lo
wer
ac
cu
r
ac
y
an
d
h
ig
h
er
laten
cy
.
SVM
in
tr
o
d
u
ce
s
lim
ited
ad
ap
tab
ilit
y
b
y
lear
n
in
g
n
o
n
li
n
ea
r
r
elatio
n
s
h
ip
s
b
etwe
en
tr
af
f
ic
f
ea
tu
r
es
an
d
Qo
S
o
u
tco
m
es;
h
o
wev
er
,
it
lack
s
tem
p
o
r
al
m
o
d
elin
g
,
w
h
ich
r
est
r
icts
its
p
er
f
o
r
m
an
ce
u
n
d
er
d
y
n
am
ic
tr
af
f
ic
p
atter
n
s
.
Fig
u
r
e
5
.
C
o
m
p
a
r
ativ
e
n
etwo
r
k
-
lev
el
p
er
f
o
r
m
a
n
ce
: static
,
SVM
an
d
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
f
r
am
ewo
r
k
(
L
STM
F
)
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
N
eu
r
a
l n
etw
o
r
k
a
p
p
r
o
a
ch
es fo
r
q
u
a
lity
-
of
-
s
ervice
o
p
timiz
a
tio
n
in
s
o
ftw
a
r
e
-
d
efin
ed
…
(
Mu
q
a
mu
d
d
i
n
Mu
h
ib
)
833
T
h
e
L
STM
-
b
ased
r
o
u
tin
g
a
p
p
r
o
ac
h
d
em
o
n
s
tr
ates
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
ac
r
o
s
s
all
m
etr
ics.
I
ts
ab
ilit
y
to
ca
p
tu
r
e
tem
p
o
r
al
d
e
p
en
d
en
cies
allo
ws
f
o
r
p
r
ed
ictiv
e
Qo
S
m
an
ag
em
en
t,
en
ab
lin
g
th
e
co
n
t
r
o
ller
t
o
p
r
o
ac
tiv
ely
r
er
o
u
te
f
lo
ws
b
ef
o
r
e
co
n
g
esti
o
n
o
cc
u
r
s
.
T
h
is
r
esu
lts
in
a
1
6
.
1
%
im
p
r
o
v
em
e
n
t
in
ac
cu
r
ac
y
o
v
er
s
tatic
r
o
u
tin
g
a
n
d
a
n
8
.
7
%
i
m
p
r
o
v
e
m
en
t
o
v
er
SVM.
E
n
d
-
to
-
en
d
laten
c
y
is
r
ed
u
ce
d
b
y
m
o
r
e
t
h
an
5
0
%
co
m
p
ar
ed
to
s
tatic
r
o
u
tin
g
,
a
n
d
th
r
o
u
g
h
p
u
t
in
cr
ea
s
es
b
y
o
v
e
r
2
5
%
r
elativ
e
to
SVM.
Pack
e
t
lo
s
s
is
m
in
im
ized
to
0
.
8
%,
r
e
f
lectin
g
e
f
f
icien
t
co
n
g
esti
o
n
a
v
o
id
an
ce
,
wh
i
le
co
n
tr
o
ller
o
v
er
h
ea
d
r
em
a
in
s
lo
w
at
8
.
3
%,
co
n
f
ir
m
in
g
th
e
m
o
d
el’
s
p
r
ac
tical
d
ep
lo
y
ab
ilit
y
.
Ov
er
all,
L
STM
-
b
ased
r
o
u
tin
g
s
ig
n
if
i
ca
n
tly
o
u
tp
e
r
f
o
r
m
s
co
n
v
en
tio
n
al
an
d
SVM
m
eth
o
d
s
,
p
r
o
v
id
in
g
b
o
th
p
r
e
d
ictiv
e
in
tellig
en
ce
an
d
ef
f
icien
t
SDN
r
eso
u
r
ce
m
an
ag
em
en
t.
Fig
u
r
es
5
an
d
6
p
r
esen
t
t
h
e
n
etwo
r
k
-
lev
el
c
o
m
p
ar
is
o
n
o
f
s
tatic
r
o
u
tin
g
,
SVM
-
b
ased
r
o
u
tin
g
,
an
d
L
STM
-
b
ased
p
r
ed
ictiv
e
r
o
u
ti
n
g
.
Static
r
o
u
tin
g
,
r
ely
in
g
o
n
p
r
ed
ef
in
e
d
p
ath
s
,
e
x
h
ib
its
th
e
h
ig
h
est
laten
cy
an
d
lo
west
th
r
o
u
g
h
p
u
t,
wh
ile
S
VM
p
r
o
v
id
es
m
o
d
er
ate
im
p
r
o
v
em
e
n
t
b
y
m
o
d
elin
g
n
o
n
lin
ea
r
tr
af
f
ic
-
Q
o
S
r
elatio
n
s
h
ip
s
b
u
t la
ck
s
tem
p
o
r
al
awa
r
en
ess
.
T
h
e
L
STM
-
b
ased
ap
p
r
o
ac
h
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
s
b
o
th
alter
n
ativ
es:
−
Pre
d
ictiv
e
in
tellig
en
ce
: te
m
p
o
r
al
m
o
d
elin
g
allo
ws p
r
o
ac
tiv
e
r
er
o
u
tin
g
b
ef
o
r
e
co
n
g
esti
o
n
o
c
cu
r
s
.
−
Per
f
o
r
m
an
ce
g
ain
s
:
a
cc
u
r
ac
y
im
p
r
o
v
es
b
y
1
6
.
1
%
o
v
er
s
t
atic
r
o
u
tin
g
a
n
d
8
.
7
%
o
v
er
SVM;
laten
cy
d
ec
r
ea
s
es b
y
m
o
r
e
th
an
5
0
%,
an
d
th
r
o
u
g
h
p
u
t r
is
es ~2
5
% a
b
o
v
e
SVM.
−
E
f
f
icien
cy
:
p
ac
k
et
lo
s
s
is
m
in
im
ized
,
p
ath
s
etu
p
tim
es
ar
e
r
e
d
u
ce
d
,
an
d
co
n
tr
o
ller
o
v
er
h
ea
d
r
em
ain
s
lo
w
at
8
.
3
%,
en
s
u
r
in
g
p
r
a
ctica
l d
e
p
lo
y
ab
ilit
y
.
Fig
u
r
e
6
co
m
p
ar
es
n
etwo
r
k
-
l
ev
el
p
er
f
o
r
m
a
n
ce
f
o
r
s
tatic
r
o
u
tin
g
,
SVM
-
b
ased
r
o
u
tin
g
,
a
n
d
L
STM
-
b
ased
p
r
ed
ictiv
e
r
o
u
ti
n
g
in
th
e
SDN
en
v
ir
o
n
m
en
t.
Static
r
o
u
tin
g
ex
h
ib
its
th
e
h
ig
h
est
laten
cy
an
d
lo
west
th
r
o
u
g
h
p
u
t
b
ec
au
s
e
it
r
elies
o
n
p
r
e
-
d
ef
in
e
d
p
ath
s
t
h
at
ca
n
n
o
t
ad
ju
s
t
to
r
ea
l
-
tim
e
n
etwo
r
k
co
n
d
itio
n
s
.
SVM
-
b
ased
r
o
u
tin
g
p
r
o
v
id
es
m
o
d
er
ate
im
p
r
o
v
em
en
ts
b
y
lear
n
in
g
n
o
n
-
lin
ea
r
tr
af
f
ic
-
Q
o
S
r
elatio
n
s
h
ip
s
b
u
t
lack
s
th
e
tem
p
o
r
al
m
o
d
elin
g
n
ec
ess
ar
y
f
o
r
p
r
ed
ictiv
e
f
lo
w
m
an
ag
em
e
n
t.
T
h
e
L
STM
-
b
ased
ap
p
r
o
ac
h
co
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
s
b
o
th
alt
er
n
ativ
es
ac
r
o
s
s
all
n
etwo
r
k
m
etr
ics.
I
ts
p
r
ed
ictiv
e
ca
p
a
b
ilit
y
allo
ws
p
r
o
ac
tiv
e
r
er
o
u
tin
g
,
r
ed
u
ci
n
g
e
n
d
-
to
-
en
d
laten
cy
b
y
m
o
r
e
t
h
an
5
0
%
co
m
p
ar
ed
t
o
s
tatic
r
o
u
tin
g
an
d
ac
h
iev
in
g
th
r
o
u
g
h
p
u
t
g
ain
s
o
f
~2
5
%
o
v
er
SVM.
Pack
et
lo
s
s
is
m
in
im
ized
to
0
.
8
%,
r
ef
lectin
g
ef
f
icien
t
co
n
g
esti
o
n
av
o
id
an
ce
.
Fas
ter
p
ath
s
etu
p
tim
es
f
u
r
th
er
im
p
r
o
v
e
r
o
u
t
in
g
ef
f
icien
cy
,
a
n
d
co
n
tr
o
ller
o
v
er
h
ea
d
r
em
ai
n
s
lo
w,
d
em
o
n
s
tr
atin
g
p
r
ac
tical
d
ep
lo
y
a
b
ilit
y
.
T
h
ese
r
esu
lts
h
ig
h
lig
h
t
L
STM
’
s
s
tr
en
g
th
in
co
m
b
in
in
g
tem
p
o
r
al
s
eq
u
en
ce
m
o
d
elin
g
with
r
e
al
-
tim
e
SDN
co
n
tr
o
l
to
o
p
tim
ize
Qo
S,
m
ak
in
g
it
h
ig
h
ly
s
u
itab
le
f
o
r
d
y
n
am
ic,
h
eter
o
g
en
eo
u
s
n
etwo
r
k
en
v
ir
o
n
m
en
ts
.
Fig
u
r
e
6
.
Netwo
r
k
-
lev
el
p
e
r
f
o
r
m
an
ce
co
m
p
ar
is
o
n
: static,
SVM,
an
d
L
STM
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
TEL
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
24
,
No
.
3
,
J
u
n
e
20
26
:
8
2
5
-
8
3
9
834
3
.
5
.
P
er
-
t
ra
f
f
ic
-
t
y
pe
a
na
ly
s
is
T
ab
le
7
f
o
cu
s
es
o
n
p
er
-
tr
a
f
f
ic
-
ty
p
e
an
al
y
s
is
,
s
h
o
win
g
h
o
w
th
e
L
STM
-
b
ased
r
o
u
tin
g
h
an
d
les
h
eter
o
g
en
e
o
u
s
f
lo
ws
in
d
i
v
id
u
ally
.
Vo
I
P
f
lo
ws,
wh
ich
a
r
e
d
elay
-
s
en
s
itiv
e,
ac
h
iev
e
th
e
h
ig
h
est
laten
cy
r
ed
u
ctio
n
(
5
5
.
1
%)
a
n
d
th
r
o
u
g
h
p
u
t
im
p
r
o
v
em
en
t
(
5
2
.
3
%),
w
h
ile
m
u
ltime
d
ia
tr
af
f
ic
b
en
ef
it
s
f
r
o
m
b
an
d
wid
th
-
awa
r
e
p
r
ea
llo
ca
tio
n
,
r
e
d
u
cin
g
laten
cy
b
y
5
3
.
8
%
an
d
in
cr
ea
s
in
g
th
r
o
u
g
h
p
u
t
b
y
5
0
.
5
%.
HT
T
P
an
d
I
o
T
f
lo
ws
also
d
em
o
n
s
tr
ate
co
n
s
is
ten
t
im
p
r
o
v
e
m
en
ts
,
m
ain
tain
in
g
p
a
ck
et
lo
s
s
b
elo
w
1
%.
T
h
is
T
ab
le
7
co
m
p
lem
e
n
ts
T
ab
le
6
b
y
p
r
o
v
i
d
in
g
tr
a
f
f
ic
-
s
p
ec
if
ic
in
s
ig
h
ts
,
h
ig
h
lig
h
tin
g
th
e
f
r
am
ew
o
r
k
’
s
ca
p
ab
ilit
y
t
o
ad
ap
t
t
o
d
iv
e
r
s
e
f
lo
w
ch
ar
ac
ter
is
tics
r
ath
er
th
a
n
ag
g
r
e
g
ated
av
e
r
ag
es.
T
ab
le
7
.
Per
-
tr
a
f
f
ic
-
ty
p
e
Qo
S
p
er
f
o
r
m
an
ce
(
L
STM
-
b
ased
r
o
u
tin
g
)
Tr
a
f
f
i
c
t
y
p
e
La
t
e
n
c
y
r
e
d
u
c
t
i
o
n
(
%)
Th
r
o
u
g
h
p
u
t
i
n
c
r
e
a
se
(
%)
P
a
c
k
e
t
l
o
ss
(
%)
H
TTP
5
1
.
2
4
8
.
7
0
.
9
V
o
I
P
5
5
.
1
5
2
.
3
0
.
7
M
u
l
t
i
me
d
i
a
5
3
.
8
5
0
.
5
0
.
8
I
o
T
5
0
.
3
4
9
.
8
1
.
0
3
.
6
.
Vis
ua
liza
t
io
n
o
f
t
hro
ug
hp
ut
a
nd
la
t
ency
Fig
u
r
e
7
p
r
esen
ts
a
d
u
al
-
ax
is
co
m
p
ar
is
o
n
o
f
th
r
o
u
g
h
p
u
t
(
Mb
p
s
,
s
h
o
wn
as
b
ar
s
)
an
d
laten
cy
(
m
s
,
s
h
o
wn
as
a
lin
e)
f
o
r
th
r
ee
r
o
u
tin
g
ap
p
r
o
ac
h
es
in
t
h
e
SDN
e
n
v
ir
o
n
m
en
t:
Static
r
o
u
tin
g
,
S
VM
-
b
ased
ad
ap
tiv
e
r
o
u
tin
g
,
an
d
L
STM
-
b
ased
p
r
ed
ictiv
e
r
o
u
tin
g
.
T
h
e
r
esu
lts
d
em
o
n
s
tr
ate
a
clea
r
p
er
f
o
r
m
a
n
ce
h
ier
ar
c
h
y
.
T
h
e
L
STM
-
b
ased
m
eth
o
d
ac
h
iev
e
s
th
e
h
ig
h
est
th
r
o
u
g
h
p
u
t
at
1
2
8
0
Mb
p
s
,
r
e
p
r
esen
tin
g
a
s
ig
n
i
f
ican
t
in
cr
ea
s
e
o
v
er
b
o
th
th
e
SVM
(
1
0
2
5
Mb
p
s
)
an
d
Static
(
8
5
0
Mb
p
s
)
b
aselin
es.
C
o
n
cu
r
r
e
n
tly
,
L
STM
m
a
in
tain
s
th
e
lo
west
av
er
ag
e
laten
c
y
at
6
7
.
4
m
s
,
in
s
tar
k
co
n
tr
ast
to
th
e
h
ig
h
er
laten
cies
o
f
SVM
(
1
0
2
.
3
m
s
)
an
d
Static
r
o
u
tin
g
(
1
4
5
.
7
m
s
)
.
T
h
is
in
v
er
s
e
r
elat
io
n
s
h
ip
wh
er
e
h
ig
h
er
th
r
o
u
g
h
p
u
t
co
in
cid
es
with
lo
wer
laten
cy
f
o
r
th
e
L
STM
ap
p
r
o
ac
h
u
n
d
er
s
co
r
es
th
e
e
f
f
i
cien
cy
o
f
p
r
ed
ictiv
e
in
tellig
en
ce
.
W
h
ile
s
tatic
r
o
u
tin
g
s
u
f
f
er
s
f
r
o
m
co
n
g
esti
o
n
d
u
e
to
n
o
n
-
ad
ap
tiv
e
p
ath
s
,
a
n
d
SVM
o
f
f
e
r
s
m
o
d
er
ate
im
p
r
o
v
em
e
n
t
b
y
r
ea
ctin
g
to
cu
r
r
en
t
co
n
d
itio
n
s
,
th
e
L
STM
m
o
d
el'
s
ab
ilit
y
to
f
o
r
e
ca
s
t
tr
af
f
ic
p
atter
n
s
en
ab
les
p
r
o
ac
tiv
e
lo
ad
b
alan
cin
g
.
T
h
is
p
r
ee
m
p
tiv
e
ac
tio
n
p
r
ev
en
ts
q
u
eu
e
b
u
ild
u
p
,
th
er
eb
y
s
im
u
ltan
e
o
u
s
ly
m
a
x
i
m
izin
g
lin
k
u
tili
za
tio
n
(
h
ig
h
th
r
o
u
g
h
p
u
t)
an
d
m
in
im
izin
g
p
ac
k
et
q
u
eu
in
g
d
elay
s
(
lo
w
laten
cy
)
.
T
h
e
v
is
u
aliza
tio
n
co
n
cr
etely
v
alid
ates
th
e
co
r
e
th
esis
th
at
n
eu
r
al
n
etwo
r
k
-
d
r
iv
en
p
r
e
d
ic
tiv
e
co
n
tr
o
l
o
p
tim
izes
m
u
ltip
le,
o
f
te
n
co
m
p
etin
g
,
Qo
S
m
etr
ics
in
tan
d
em
,
lead
in
g
to
a
s
u
p
er
io
r
o
v
er
all
n
etwo
r
k
s
tate
co
m
p
ar
e
d
to
co
n
v
en
tio
n
al
o
r
s
h
allo
w
lear
n
in
g
te
ch
n
iq
u
es.
Fig
u
r
e
7
.
T
h
r
o
u
g
h
p
u
t
an
d
laten
cy
p
er
f
o
r
m
a
n
ce
co
m
p
ar
is
o
n
f
o
r
s
tatic
,
SVM,
an
d
L
STM
-
b
a
s
ed
r
o
u
tin
g
ap
p
r
o
ac
h
es
3
.
7
.
Abla
t
io
n
s
t
ud
y
,
g
ener
a
li
za
t
io
n,
a
nd
s
t
a
t
is
t
ica
l r
o
bu
s
t
nes
s
T
o
r
ig
o
r
o
u
s
ly
v
alid
ate
th
e
r
el
iab
ilit
y
,
r
o
b
u
s
tn
ess
,
an
d
d
e
p
lo
y
ab
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
L
STM
-
b
ased
Qo
S
o
p
tim
izatio
n
f
r
am
ew
o
r
k
,
we
co
n
d
u
cte
d
a
co
m
p
r
eh
en
s
iv
e
ev
alu
atio
n
alo
n
g
f
iv
e
d
im
en
s
io
n
s
:
s
tatis
tical
s
ig
n
if
ican
ce
,
h
y
p
e
r
p
ar
am
eter
ab
latio
n
,
to
p
o
lo
g
y
g
en
e
r
aliza
tio
n
,
r
esil
ien
ce
u
n
d
er
d
y
n
a
m
ic
co
n
d
itio
n
s
,
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.