Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
8
,
No.
6
,
D
ece
m
ber
201
8
, pp.
4467
~
44
76
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v8
i
6
.
pp4467
-
44
76
4467
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Effecti
ve
Route
r
Assisted
Congestion C
ont
ro
l
f
or S
DN
So
fi
a
Naning
Hertia
na
1
,
A
d
it Kur
niawa
n
2
, H
en
draw
an
3
,
U
d
jianna
Sek
teria P
asari
bu
4
1
,2,3
School
of El
e
ct
ri
ca
l
Eng
ine
er
i
ng
and
In
form
at
i
cs
,
Insti
tut Te
kn
ologi
B
andung, I
ndonesia
4
Facul
t
y
of
Ma
th
emati
cs
and
Na
t
ura
l
Sc
ie
nc
es,
In
stit
ut Te
kno
logi
Bandung
,
Indon
esia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
r 16
, 201
8
Re
vised
Ju
l
6
,
201
8
Accepte
d
J
ul
20
, 2
01
8
Route
r
As
sisted
Congesti
on
Con
trol
(RACC)
wa
s
designe
d
to
improve
end
-
to
-
end
conge
st
io
n
cont
rol
per
for
m
anc
e
b
y
using
prior
knowledge
on
net
work
condi
ti
on
.
How
eve
r,
the
tra
d
i
ti
onal
Internet
does
not
pr
ovide
such
informati
on,
whi
ch
m
ake
s
thi
s
ap
proa
ch
is
not
f
eas
ibl
e
to
de
li
ve
r.
Our
pape
r
addr
esses
thi
s
n
et
work
informat
ion
def
i
cienc
y
i
ss
ue
b
y
proposi
ng
a
n
ew
conge
stion
control
m
et
hod
th
at
works
on
the
Software
Defi
n
e
d
Network
(SD
N)
fra
m
ew
ork.
W
e
ca
ll
thi
s
proposed
m
et
hod
as
PACE
C
(Path
As
socia
ti
vity
C
e
ntra
liz
ed
Cong
e
stion
Control
).
I
n
SD
N,
globa
l
vie
w
of
the
net
work
informati
on
cont
a
ins
the
net
work
topol
og
y
including
li
n
k
prope
rti
es
(i.
e
.
,
t
y
pe,
ca
p
acit
y
,
power
consum
pti
on,
etc.
)
.
PA
CEC
uses
thi
s
informati
on
to
det
ermine
the
fee
dbac
k
signal,
in
orde
r
for
the
s
ourc
e
to
start
sending
data
at
a
high
r
ate
an
d
to
quic
kl
y
r
each
fai
r
-
shar
e
ra
t
e.
Th
e
sim
ula
ti
o
n
show
s
tha
t
the
eff
ic
i
ency
a
nd
fai
rne
ss
of
PA
CEC
are
be
tt
e
r
tha
n
Tr
ansm
ission
Control
Protocol
(
TCP)
and
Ra
te Cont
ro
l
Protoco
l
(
RCP
)
.
Ke
yw
or
d:
C
o
n
g
e
s
t
i
o
n
c
o
n
t
r
o
l
G
l
o
b
a
l
k
n
o
w
l
e
d
g
e
P
A
C
E
C
S
D
N
T
r
a
d
i
t
i
o
n
a
l
I
n
t
e
r
n
e
t
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
So
fia
N
a
ning
Her
ti
ana
,
School
of Elec
tric
al
Engineer
ing
a
nd
Inform
at
ic
s,
In
sti
tut Te
knol
og
i
Ba
ndun
g,
I
ndonesi
a
,
J
l.
Ga
nes
ha No.1
0,
Ba
ndun
g 4
0132, I
ndonesi
a
.
Em
a
il
:
so
fiana
ning
@
te
lk
om
un
ive
rsity
.ac.id
1.
INTROD
U
CTION
Congesti
on
co
ntr
ol
is
a
proc
ess
to
re
gu
la
te
the
sen
ding
r
at
e
of
the
s
our
ce
on
t
he
net
work
so
t
he
sen
der
m
ay
adj
ust
the
entry
of
data
acc
ordin
g
t
o
the
netw
ork
co
ndit
ion
to
ens
ure
se
nder'
s
Q
ualit
y
of
Se
rv
ic
e
(QoS
)
re
quire
m
ents
are
f
ulfi
ll
ed
[1
]
.
A
go
od
c
onge
sti
on
con
t
ro
l
will
ensu
re
t
hat
there
is
no
dro
p
in
qu
al
it
y
and
ass
ure
a
good
qual
it
y
network
[2
]
,
[
3].
I
t
is
widel
y
kn
own
that
a
n
en
d
-
to
-
e
nd
pac
ket
loss
co
ntr
ol
suc
h
as
Transm
issi
on
Con
tr
ol
P
r
oto
c
ol
(TC
P)
has
s
ever
al
i
neffici
encies
[
4].
O
ne
reas
on
of
i
neffici
ency
is
that
TCP
treat
s
pac
ket
lo
ss
as
co
ngest
io
n
sig
nal.
T
here
fore,
it
can
not
disti
nguish
bet
ween
pac
ket
lo
ss
due
to
c
onge
sti
on
with
pack
et
l
oss d
ue
t
o other c
auses.
Packet
loss
a
s
a
congesti
on
si
gn
al
im
plies
that
act
ion
can
only
be
do
ne
af
te
r
co
ng
est
i
on
occurs
[5
]
.
TCP
us
es
ass
um
pt
ion
that
t
he
netw
ork
doe
s
not
pro
vid
e
exp
li
ci
t
fee
db
a
ck
to
t
he
s
our
ce
[6
]
,
w
hich
m
akes
each
s
ource
to
est
i
m
at
e
the
na
ture
of
t
he
net
work
pat
h,
s
uc
h
as
r
ound
tri
p
tim
e
(RTT)
or
us
a
ble
ba
ndwi
dth
,
t
o
deliver
e
ff
ic
ie
nt
end
-
to
-
en
d
c
ongestio
n.
To
i
m
pr
ove
TCP
perform
ance,
previ
ou
s
resea
rc
h
intr
oduce
d
t
he
us
e
of
e
xp
li
ci
t
co
ng
e
sti
on
sig
na
ls
to
t
he
netw
ork
[
7].
This
m
et
ho
d
is
co
m
m
on
ly
cal
led
the
R
oute
r
A
ssist
ed
C
ongestio
n
Co
ntr
ol
(RACC).
In
the
tradit
io
nal
In
te
r
net
,
RACC
m
et
ho
d
is
app
li
ed
to
ea
ch
router
wh
e
r
e
the
router
pro
vid
es
feedbac
k
to
t
he
end
syst
em
a
bout
the
sta
te
of
the
netw
ork
a
nd
te
ll
s
t
he
se
nder
t
o
se
nd
pac
kets
at
a
spe
ci
fic
rat
e.
Fi
gure 1
g
iv
es
a
sim
ple
ov
erv
ie
w
of
r
oute
rs
th
at
al
ways p
r
ovide f
ee
db
a
ck
i
nfor
m
at
ion
to
t
he
end
-
us
er
.
RAC
C
is
m
od
el
ed
us
in
g
the
M/
G
/1
-
PS
queue
t
he
or
y
to
cal
cula
te
the
aggreg
at
e
rate
on
eac
h
router
(no
de)
.
T
he
a
ggre
gated
r
at
e a
t eac
h node
l
f
ollows a sim
ple m
od
el
as d
es
cribe
d by the
f
ollow
i
ng equat
ion
[
8]:
=
(
1
−
)
(1)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2018
:
4467
-
4476
4468
w
he
re
R
l
is
the
sen
ding r
at
e
f
or a li
nk,
C
l
is
th
e
li
nk
ca
pacit
y,
and
ρ
l
is
t
he
li
nk u
ti
li
zat
ion
.
Ther
e
a
re
thr
e
e
so
luti
ons
off
ered
for
RAC
C
in
the
tradit
ion
al
I
nter
net.
The
fir
st
so
luti
on
is
to
us
e
a
router
to
detec
t
congesti
on
a
nd
se
nd
this
in
form
ation
to
t
he
en
d
syst
em
.
Ba
sed
on
this
inform
at
ion
,
the
en
d
syst
e
m
decides
to
c
on
tr
ol
the
congesti
on
on
t
he
netw
ork.
Th
e
exam
ple
of
t
his
cat
eg
or
y
is
Ex
plici
t
Con
ge
sti
on
No
ti
ficat
io
n
(
ECN)
[
9]
an
d
Qu
ic
k
-
Start
[
10
]
.
The
sen
di
ng
rates
in
bo
th
prot
oco
ls
ar
e
decide
d
at
t
he
e
nd
syst
e
m
based
on
i
nfor
m
at
ion
recei
ved
f
rom
the
netw
ork
and
the
c
har
a
ct
erist
ic
s
of
ea
ch
a
p
plica
ti
on.
The
seco
nd
s
olu
ti
on
is
to
us
e
a
router
to
detect
congesti
on
an
d
avail
able
netw
ork
res
ources
,
and
at
the
sam
e
tim
e,
the
router
distr
ibu
te
s
netw
ork
resour
ce
s
f
or
each
inf
or
m
at
i
on
fl
ow.
The
e
xam
ple
of
this
cat
ego
ry
is
Exp
li
ci
t
Congesti
on
Co
ntr
ol
Pr
oto
c
ol
(X
CP
)
[11],
a
nd
Ra
te
Co
ntr
ol
Protoc
ol
(RC
P)
[12].
I
n
t
his
seco
nd
ap
proa
ch,
t
he
end
syst
em
on
ly
acce
pts r
eco
m
m
end
at
ion
s fro
m
the
r
oute
r
to
ad
just
it
s
se
nd
i
ng
rate.
T
his
ap
proac
h
al
lo
ws
t
he
router t
o decid
e the se
ndin
g
r
at
e for
eac
h flo
w wit
hout ca
usi
ng
c
onge
sti
on
on the
netw
ork
.
Figure
1
.
Ro
uter
assist
ed
con
gestio
n
c
on
tr
ol
To
cal
c
ulate
the
c
hange
i
n
flo
w
se
nd
i
ng
r
at
e,
XCP
cal
culat
es
the
a
gg
reg
at
e
ba
ndwi
dth
f
or
eac
h
router.
X
C
P
use
s the
fo
ll
owin
g
e
qu
at
io
n
t
o
c
al
culat
e the
de
sired
a
djust
m
e
nt of t
he
a
ggre
gate b
a
ndwi
dth [
11
]
.
(
)
=
(
−
(
)
)
−
(
)
(2)
In
t
his
eq
uatio
n,
C
l
is
the
ca
pa
ci
ty
of
the o
ut
go
i
ng
li
nk,
(
)
is
t
he
rate of
it
s outg
oing
tra
ff
ic
,
(
)
is
the
pe
rsiste
nt
qu
e
ue
durin
g
t
he
pr
e
vious
c
ontr
ol
inter
val
a
nd
d
is
t
he
a
ve
rag
e
RT
T.
T
he
res
ulted
ag
gr
egate
feedbac
k
(
)
can
be
posit
ive
or
ne
gative
an
d
is
distrib
ute
d
am
on
g
the
tr
aver
si
ng
fl
ow
s
.
Th
e
fai
rn
ess
con
t
ro
ll
er
us
e
s
the
A
ddit
ive
-
I
ncr
ease/
M
ulti
plica
ti
ve
-
D
ecrea
se
(
A
IMD)
pr
i
nciple
t
o
al
loc
at
e
the
po
sit
iv
e
or
ne
gative
fee
db
ack.
Po
sit
ive
feedbac
k
is
distribu
te
d
e
qu
a
ll
y
a
m
on
g
al
l
flo
ws
a
nd
ne
gative
feedba
ck
i
s
distrib
uted
pro
portio
nally
to
their
cu
rr
e
nt
th
rou
ghput.
XC
P
has
tw
o
disa
dv
a
ntage
s:
firs
t,
the
sta
rtup
flow
is
work
i
ng
slow
l
y,
so
the
com
pleti
on
tim
e
of
flo
w
is
no
t
sm
oo
t
h
f
or
sm
al
l
flo
w
[13].
Sec
ond,
it
require
s
per
-
pack
et
c
al
culat
ion
; t
his
r
ai
ses
the pr
ob
le
m
o
f si
gn
ific
a
nt
ov
e
rh
ea
d [12].
RC
P
was
de
ve
lop
e
d
by
Na
ndit
a
[
12
]
with
the
ai
m
of
pro
vid
i
ng
a
si
m
pler
co
nge
sti
on
c
ontr
ol
m
echan
ism
. R
CP assum
es th
at
the se
nd
er
use
s a r
at
e
-
base
d
delive
ry
m
ec
han
ism
. Eq
ual
to X
CP, RC
P requ
i
res
cal
culat
ing
the
aggre
gate rate to adju
st t
he
f
l
ow
se
ndin
g
rat
e. RCP cal
culat
es the ag
gre
ga
te
r
at
e
(
)
on
ce
pe
r
interval c
ontr
ol w
it
h
t
he follo
wing e
qu
at
io
n:
(
)
=
(
−
)
(
1
+
(
−
(
)
)
−
(
(
)
)
)
(3)
Wh
e
re
(
)
is
the
com
m
on
feedba
ck
rate,
d
is
th
e
aver
a
ge
RTT
and
C
l
is
the
capaci
ty
of
the
ou
t
go
i
ng
li
nk
.
(
)
is
the
ag
gr
e
gate
ou
t
go
i
ng
t
raffic
w
hich
was
m
easur
ed
duri
ng
t
he
la
st
co
ntro
l
inter
val,
(
)
is
the
cu
rr
e
nt
qu
e
ue
occ
upat
ion
d
is
the
upda
te
interval
dur
at
ion
with
d
.
It
can
be
sho
wn
that
RC
P
is
loc
al
ly
sta
ble
if
certai
n
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
Eff
ect
iv
e Ro
ute
r Assiste
d C
on
gestio
n
Co
ntr
ol
for SD
N
(
So
fi
a
N
an
i
ng H
ert
i
ana
)
4469
conditi
ons
f
or
α
a
nd
β
a
re
fu
l
fill
ed
[
12]
.
Des
pite
thes
e
i
m
pr
ovem
ents,
XCP
an
d
RC
P
sti
ll
face
m
any
chall
enges
e
s
pe
ci
al
ly
in
que
ue
sup
port.
The
pro
blem
of
t
he
que
ue
s
uppo
rt
in
this
syst
e
m
is
that
the
prot
oco
l
assum
es
on
ly
a
sing
le
queue
,
con
tra
ry
to
th
e
desig
n
of
a
hi
gh
-
en
d
r
ou
te
r
that
rar
el
y
has
on
ly
one
que
ue
.
The
i
m
ple
m
entat
io
n
can
be
tric
ky
if
there
are
thousa
nd
s
of
f
lows
pa
ssi
ng
thr
ough
the
r
oute
rs
with
different
char
act
e
risti
cs.
Finall
y,
the
th
ird
s
olu
ti
on
is
to
us
e
a
r
oute
r
to
detect
congesti
on
al
ong
t
he
path
a
nd
pro
vide
inf
or
m
at
ion
to
the
en
d
syst
e
m
.
The
exam
ple
of
this
cat
eg
ory
is
the
Op
en
Box
P
ro
t
oco
l
(
OBP)
[14].
OB
P
us
e
s
a
colla
borati
on
ro
ute
r
to
iden
ti
fy
netwo
r
k
r
eso
ur
ces
al
on
g
the
path
an
d
deliver
this
in
f
or
m
at
ion
to
the
end
syst
e
m
.
The
e
nd
syst
em
m
a
de
the
co
ngest
ion
c
ontrol
de
ci
sion
us
in
g
t
he
data
recei
ve
d
f
r
om
the
rou
te
r.
T
he
fo
ll
owin
g
eq
ua
ti
on
sho
w
s
ho
w
the
se
nd
i
ng
r
a
te
is
adjusted
.
In
O
BP,
T
he
init
ia
l
send
in
g
rate
W
(
t
0
)
de
pe
nd
s
on
the av
ai
la
ble bandwidt
h
AB
(
t
0
),
t
he
ca
pacit
y
CB
(
t
0
)
at
the
narr
ow li
nk and
the consta
nts
α
an
d
ß
[14].
(
0
)
=
∗
(
0
)
+
∗
(
0
)
(
4)
Ever
y
tim
e
a
new
ACK
pac
ket
is
received
,
the
feedbac
k
inform
ation
inside
the
pac
ke
t
is
us
ed
to
m
ake
adjustm
ents
in
tran
sm
issi
on
rate.
O
BP
cl
aim
s
that
it
s
co
m
pu
ta
tio
n
is
sim
pler
than
XCP
an
d
RC
P.
Howe
ver,
the
send
e
r
on
OBP
sh
oul
d
create
de
ci
sion
of
ad
j
ust
in
g
the
sen
di
ng
rate
us
in
g
only
the
inform
at
ion
on
the
r
ou
te
r
al
ong
t
he
path.
This
inf
or
m
at
ion
is
sti
ll
a
local
cat
e
gory
beca
us
e
t
he
recipient
does
not
unde
rstan
d
the
act
ual
net
wor
k
c
onditi
ons.
C
on
s
eq
ue
ntly
,
th
e
OBP
sho
uld
al
ways
be
care
fu
l
i
n
inc
reasin
g
a
nd
decr
easi
ng the
delivery
rate t
o m
a
intai
n
the
ne
twork
stabil
it
y.
In
g
e
ne
ral, RA
CC
can
i
m
pr
ove n
et
work
perf
or
m
ance.
How
ever, the RACC
m
e
tho
d
is r
un b
y usin
g
a
distrib
uted
fr
a
m
ewo
r
k
as
in
tradit
ion
al
networks
has
som
e
dr
aw
bac
ks
.
It
requires
a
r
ou
te
r
(s
)
that
s
upport
s
sign
al
in
g band
width
of
t
he
se
t of
p
a
ram
et
ers
of
the
sen
ding
r
at
e [15]. T
he prese
nce
of
i
nc
om
plete
inf
orm
at
ion
about
netw
ork
co
nd
it
io
ns
m
akes
t
he
iss
ue
of
ef
fici
ency
a
nd
sta
bili
ty
.
T
he
c
ongestio
n
con
t
ro
l
schem
e
with
netw
ork
s
uppo
rt
bec
om
es
ineff
ic
ie
nt
s
o
t
hat
a
global
in
for
m
at
ion
pro
vide
r
m
echan
ism
is
nee
ded
that
can
be
us
e
d
by
co
nge
sti
on
c
on
tr
ol
m
echan
ism
s
t
o
co
ntr
ol
net
work
c
onge
sti
on.
Co
ngest
io
n
co
ntr
ol
us
i
ng
gl
ob
al
inf
or
m
at
ion
ha
s
bee
n
pro
posed
by
Mo
nia
et
al
.
[
16]
pro
posed
O
pe
nT
CP,
a
TCP
adjus
tm
ent
dynam
ic
s
fr
am
ewo
r
k
bas
ed
on
S
DN.
T
he
sen
ding r
at
e adju
stm
ent
in O
pe
nTCP
is
gl
ob
al
-
base
d
i
nfor
m
at
ion
m
anag
ed
by
the
co
ntr
oller.
Op
e
nTCP
ha
s
fill
ed
the
S
DN
a
ppli
cat
ion
ga
ps
in
t
he
congesti
on
c
on
t
ro
l
fiel
d.
H
ow
e
ve
r,
Op
e
nTCP
is
a
ne
w
ar
chite
ct
ur
e
an
d
nee
ds
sever
al
m
od
ifi
cat
ion
s
on
s
om
e
el
e
m
ents
of
the
net
work
su
c
h
as
hav
i
ng
to
m
od
ify
the
so
urce,
f
orwardin
g
no
de
s
and
co
ntr
ollers.
O
pe
n
TCP
al
so
has
no
sp
e
ci
fic
ta
rg
et
rate
an
d
adjustm
ent
m
et
hods
.
L
i
ngyun
et
al
.
[
17
]
pr
ese
nt
m
ult
iple
act
ive
qu
eue
m
anag
em
ent
al
gorithm
s.
The
al
gorithm
is
ex
ecuted
acc
ordi
ng
t
o
the
locat
i
on
of
t
he
co
ng
est
ion
on
the
ne
twork
.
This
al
gorithm
is
adap
ti
ve
to li
nk con
diti
on.
The
li
nk
c
ondi
ti
on
is
detect
ed
by
m
on
it
or
in
g
the
sta
ti
sti
cal
center,
if
ther
e
is
a
con
gestion
li
nk,
the
n
the
inf
orm
ation
fl
ow
on
th
is
li
nk
is
tra
nsfer
red
by
the
Op
e
nF
l
ow
c
on
trolle
r.
T
his
m
et
ho
d
cl
ai
m
s
to
be
eff
ect
ive
for
c
onge
sti
on
c
ontr
ol.
Ne
ve
rtheles
s,
this
al
gorith
m
do
es
no
t
fu
ll
y
ta
ke
into
acc
ount
the
c
ondi
t
ion
of
netw
ork
globa
ll
y.
Yao
[
18]
pro
po
se
d
a
n
a
lgori
thm
called
S
of
t
war
e
-
De
fine
d
Co
ngest
ion
C
ontrol
(SDCC
).
This
a
ppr
oach
has
t
he
c
har
a
ct
erist
ic
s
of
ce
ntrali
zed
c
ontr
ol
an
d
ca
n
get
a
gl
ob
al
t
opol
og
y
f
or
inte
grat
ed
netw
ork
m
anag
em
ent.
SD
CC
can
op
ti
m
ize
li
nk
util
iz
at
i
on
to
c
on
t
ro
l
netw
ork
co
nge
sti
on
.
Howe
ve
r,
SD
CC
sti
ll
on
ly
co
ns
iders
net
work
pe
rfor
m
ance
an
d
do
es
not
pay
at
te
ntion
to
fl
ow
perform
ance.
Co
ng
e
sti
on
c
on
t
ro
l
us
in
g
S
D
N
a
ppr
oach
is
sti
ll
in
early
de
velo
pm
ent
sta
ge.
T
he
pro
po
sed
wor
ks
ha
ve
no
t
discu
s
sed
the
weaknesse
s,
a
nd
pe
rfor
m
anc
es
i
m
pr
ovem
e
nts
of
router
a
ssist
ed
co
ntr
ol
in
tra
diti
on
al
netw
orks.
The
m
os
t
extensi
ve work
for
t
he fie
ld
of co
ng
e
sti
on contr
ol in
t
he
S
D
N
is
pro
po
s
ed
f
or the
data ce
nter
as i
n [19]
-
[
21]
.
In
this
pa
per,
we
pro
posed
a
new
Ro
uter
A
ssist
ed
Conges
ti
on
Co
ntro
l
(
RACC
)
m
echa
nism
t
hat
we
cal
l
Path
A
ss
ociat
ivit
y
Ce
ntrali
zed
E
xp
li
c
it
Congesti
on
Con
tr
ol
(PACEC
).
P
ACEC
works
on
the
SDN
fr
am
ewo
r
k
t
o
ov
e
rc
om
e
the
weaknesse
s
of
the
R
ACC
in
tradit
io
nal
I
nt
ern
et
by
pro
vid
in
g
global
ne
twor
k
inf
or
m
at
ion
. W
e u
se c
om
prehensi
ve
in
for
m
at
ion
to
i
m
p
ro
ve
t
he
accu
ra
cy
o
f
fee
db
a
ck
to
determ
ine t
he
sou
rce
sen
ding
rate.
By
us
ing
S
D
N
te
chnolo
gy,
c
ongestio
n
co
nt
ro
l
util
iz
es
the
global
knowle
dg
e
of
t
he
net
work
i
n
it
s
decisi
on
m
akin
g.
Data
m
on
it
ori
ng
c
ollec
ts
netw
ork
i
nfo
rm
ation
,
a
nd
this
in
form
at
i
on
is
use
d
by
the
con
t
ro
ll
er
to
m
ake
centrali
zed
decisi
ons
i
n
res
pons
e
to
changin
g
netw
ork
co
ndit
ion
s
[22].
This
ap
proac
h
pro
du
ces
ac
cu
r
at
e
inform
at
io
n
s
o
that
the
se
nd
e
r'
s
sen
ding
rate
can
be
c
ust
omi
zed
ap
pro
pr
ia
te
ly
accor
di
ng
to
netw
ork
c
ondit
ion
s
.
PA
CEC
cal
c
ulate
s
the
r
at
e
by
involvin
g
al
l
the
nodes
al
ong
the
c
onne
ct
ion
path
th
rou
gh
w
hich
th
e
inf
or
m
at
ion
flo
ws.
T
he
refor
e
,
the
sen
ding
ra
te
of
the
i
nform
at
ion
flo
w
does
no
t
need
t
o
cha
nge
as
lo
ng
a
s
it
sti
ll
passes
thr
ough
the
sam
e
path
a
nd
the
update
ti
m
er
of
the
co
ntr
oller
has
no
t
e
nd
e
d.
PA
CEC
ca
n
a
lso
set
the
source
se
ndin
g
rate
sta
rting
at
hi
gh
-
s
pe
ed
so
that
ne
twork
res
ource
s
can
be
use
d
m
or
e
eff
ic
ie
ntly
.
This
pro
po
se
d
m
echan
ism
is
the
novelty
of
this re
search
i
n
the
c
ongestio
n
c
on
t
ro
l do
m
ai
n.
Th
e
rest
of
this
pa
p
er
i
s
structu
re
d
as
f
ollows:
sect
io
n
2
e
xpla
ins
t
he
relat
ed
w
orks
within
t
he
t
op
i
c
of
the
RACC
.
Sect
io
n
3
des
cribes
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2018
:
4467
-
4476
4470
the
desig
n
of
the
pro
po
se
d
m
echan
ism
in
this
pap
er
.
S
ect
ion
4
de
scr
ibes
the
si
m
ulati
on
res
ults
of
the
pro
po
se
d
m
echan
ism
, an
d fina
ll
y, Sect
i
on
5 d
escribes
the c
oncl
us
i
on
s
and
po
s
sible
fu
t
ur
e
r
esea
rch dire
ct
ion
s
.
2.
PROP
OSE
D MET
HO
D
This
sect
ion
prov
i
des
an
ove
r
view
of
the
P
ACEC
dr
a
ft
[23].
To
bette
r
unde
rstan
d
the
m
ot
ivati
on
s
beh
i
nd
PA
CE
C,
le
t
us
rem
e
m
ber
that
RACC
us
es
the
ba
sic
equ
at
io
ns
we
ha
ve
wr
it
te
n
in
E
quat
ion
(1)
t
o
determ
ine the
aggre
gate rate
on a lin
k.
F
r
om
that equ
at
io
n we ca
n
see t
ha
t i
nf
orm
at
ion
of
a
vaila
ble
ba
ndwidt
h
cov
e
rs
only
w
it
hin
local
.
Co
ns
e
qu
e
ntly
,
an
RACC
router
will
need
to
coor
din
at
e
with
oth
e
r
r
oute
rs
wh
e
n
i
m
ple
m
ented
i
nto
netw
orks
with
m
ulti
ple
ro
ute
rs.
T
o
a
void
this
draw
bac
k
,
PA
CEC
im
ple
m
ents
a
centr
al
iz
e
d
congesti
on
co
ntr
ol
m
echan
is
m
,
wh
il
e
the
RACC
router
on
t
he
tradit
io
nal
In
te
rn
et
ca
lc
ulate
s
the
ag
gr
e
gate
rate
f
or
a
li
nk
on
ly
.
PA
CEC
can
cal
c
ulate
the
a
ggre
gate
r
at
e
for
a
pat
h,
wh
e
re
t
he
path
has
bee
n
pro
vi
ded
t
o
distrib
ute
the
f
low
from
the
so
urce
to
the
de
sti
nation.
W
it
h
this
m
echan
ism
,
the
flow
is
gu
a
ra
nteed
by
route
and
rate
w
he
n
the
flow
is
al
lowed
to
e
nter
into
the
net
work.
O
ur
pr
opose
d
m
et
ho
d
is
descr
ibed
i
n
t
he
fo
ll
owin
g
sect
i
on
s
.
2.1.
Pa
th
R
at
e
(
R
p
)
Si
m
il
ar
to
the
case
of
the
R
ACC
r
ou
te
rs
,
PA
CEC
requir
es
cal
culat
i
on
on
ag
gregate
r
at
es
to
a
dju
st
the
sen
ding
ra
te
.
The
dif
fer
e
nce
is
that
the
RACC
cal
culat
es
an
aggre
ga
te
rate
fo
r
one
li
nk
only
.
PA
CEC
cal
culat
es
the
aggre
gate
rate
for
a
pat
h.
W
e
cal
l
agg
re
gate
rate
on
PA
CE
C
as
path
rate
(
).
To
get
the
pat
h
rate,
we
co
ns
id
er
a
netw
ork
m
od
el
w
hose
to
polo
gy
is
char
ac
te
rized
in
Fig
ure
2.
T
he
netw
ork
m
od
el
is
a
path
consi
sti
ng
of
s
ever
al
li
nk
s
(
1,
2,
...,
H
).
T
he
wh
ole
li
nk
is
connecte
d
to
a
c
entrali
zed
co
nt
ro
ll
er.
E
nd
syst
e
m
as
the
source
of
inf
or
m
at
ion
flo
w
is
connecte
d
to
the
ing
res
s
switc
h
as
the
ga
te
way
to
the
netw
ork.
I
n
thi
s
case,
nodes
C
1
se
rves
as
in
gr
es
s
s
witc
h.
T
he
s
ourc
e
f
i
has
a
n
a
ssoc
ia
te
d
sen
ding
rate
o
f
r
i
.
Flo
w
is
transm
it
te
d
from
so
urce t
o desti
nation via
a
path
that
has
capa
ci
ty
C
p
.
Figure
2
.
Net
w
ork
m
od
el
of
P
ACEC
A path c
onta
in
ing
a
ser
ie
s
of
H
li
nk(s
) wil
l h
ave th
e
p
at
h rate
(
R
p
)
for
m
ulate
d by the
f
ollo
wing e
qu
at
io
n:
=
min
=
1
…
.
(
1
−
)
(5)
Γ
i
is
the
i_
th
a
ver
a
ge
li
nk
util
iz
at
ion
of
colle
ct
ed
by
co
ntr
ol
le
r
i
,
w
her
e
=
+
,
y
i
is
the
i
_
th
ave
rag
e
tra
ff
i
c
li
nk
,
q
i
is
the
i_
th
qu
e
ue
li
nk,
C
i
is
the
i_
th
li
nk
capaci
t
y.
In
co
ntrast
to
RACC
in
t
he
tra
diti
on
al
In
te
r
net,
PA
CEC
us
e
s
Γ
i
,
w
hic
h
is
a
util
iz
at
ion
m
atr
ix
w
ho
se
el
e
m
ents
are
glob
al
inf
or
m
at
ion
netw
ork
.
Inform
ation
about
Γ
i
is
obt
ai
ned
by
the
c
on
t
ro
ll
er
globa
ll
y
fr
om
each
switc
h
inc
orp
orat
ed
in
the
c
on
trolle
d
netw
ork
.
T
he
r
i
Inf
orm
ation
of
R
p
New control
action
C
1
C
2
Co
n
t
r
o
l
l
er
:
Co
n
g
e
s
t
i
o
n
co
n
t
ro
l
p
o
l
i
cy
So
u
rce
f
i
D
es
t
i
n
at
i
o
n
C
H
C
p
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
Eff
ect
iv
e Ro
ute
r Assiste
d C
on
gestio
n
Co
ntr
ol
for SD
N
(
So
fi
a
N
an
i
ng H
ert
i
ana
)
44
71
m
ai
n
adv
a
ntag
e
of
is
that
it
can
be
a
pr
e
ci
se
so
luti
on
f
or
,
w
hi
ch
is
no
t
under
-
est
i
m
at
e
d
n
or
ove
r
-
est
i
m
at
e
d
.
Sub
-
sect
ion 3
.
2
de
scribes
ho
w
P
ACEC
ob
ta
ins
su
c
h
in
form
at
i
on.
2.2.
R
p
Up
d
at
i
ng
Netw
ork
c
ondi
ti
on
s
var
y
fro
m
tim
e
to
tim
e.
Th
ere
fore
t
he
c
on
t
ro
ll
er
ne
eds
to
up
date
the
netw
ork
inf
or
m
at
ion
co
ntinuo
us
ly
to
de
li
ver
an
ef
fecti
ve
poli
cy
.
T
he
co
ntro
ll
er
upda
te
s
the
inf
orm
at
ion
of
R
p
in
e
ach
update
pe
rio
d
T
c
an
d
the
c
ontrolle
r
will
pro
vid
e
updated
i
nfor
m
at
ion
to
the
in
gr
ess
swit
ch.
R
p
is
af
fect
ed
by
the
switc
h
util
i
zat
ion
Γ
i
.
R
p
ch
ang
e
s
if
t
her
e
is
a
cha
nge
in
Γ
i
.
I
n
t
his
sect
ion,
we
descr
i
be
how
the
c
ontrolle
r
ob
ta
in
s
the
up
dated
i
nfor
m
ation
on
.
Sup
pose
there
are
H
switc
h
es
t
hat
s
end
the
update
d
in
form
at
ion
to
th
e
con
t
ro
ll
er
i
n
c
on
sta
nt
ti
m
e
interval
t
.
T
hus
,
each
delive
ry
of
data
is
done
at
=
(
∆
,
2
∆
,
3
∆
,
…
…
.
∆
)
,
wh
e
re
T
s
is
ti
m
e
update
s
witc
h.
T
he
c
ontr
oller
processes
the
in
f
or
m
at
i
on
f
or
the
c
on
trolle
r
at
a
pe
r
iod
T
c
,
with
=
.
so
that
ev
ery
tim
e
the
up
date
is
pe
rfo
r
m
ed,
the
co
ntr
oller
colle
ct
s
inf
or
m
at
ion
m
t
i
m
es
with
interval
T
s
.
I
f
the
total
ob
se
rvat
ion
tim
e
is
T
and
durin
g
this
tim
e
interval
t
her
e
are
k
ti
m
e
s
wh
e
re
inf
or
m
at
ion
is
colle
ct
ed
,
then
the
co
ntr
oller
requires
an
up
date
of
tim
es
that
are
done
at
tim
e
interval
ind
e
x
s
1
,
s
2
,
…
.
s
k
m
−
1
,
s
k
m
,
wh
e
re
is
tota
l
inf
or
m
at
io
n
a
t
obser
vatio
n
ti
m
e
T
as
s
how
n
i
n
Fig
ur
e
3
.
Her
e
s
c
de
no
te
s
the
in
dex of ti
m
e interval
where a c
ontrolle
r
co
ll
ect
s
in
for
m
at
ion
f
r
om
each
s
witc
h.
j
=
0
j
=
4
j
=
m
k
=
1
k
=
2
k
=
m
+
1
k
=
s
.
m
k
=
m
+
3
T
j
=
2
j
=
m
+
2
j
=
m
+
4
j
=
m
+
4
s
1
s
2
s
k
/
m
T
c
=
m
.
Ts
Ts
Figure
3
.
I
nf
or
m
at
ion
updat
e
Link uti
li
zat
ion
in
each
ti
m
e
ind
e
x
s
c
for
eac
h
li
nk
i
ca
n be
cal
culat
ed
as
(
)
=
1
∑
,
+
(
−
1
)
+
,
+
(
−
1
)
)
=
1
(
6)
Wh
e
re
,
repre
sents
the
pac
ket
passi
ng
th
rou
gh
a
switc
h
i
at
ti
m
e
ind
ex
(
=
.
∆
)
,
wh
e
reas
,
represe
nts
the
qu
e
ue
le
ngth
a
t
switc
h
i
at
(
=
.
∆
)
.
The
y
i
an
d
q
i
a
re
c
ollec
te
d
e
ve
ry
tim
e
interv
al
Ts
t
o
k.Ts
.
The
y
i,p
is
the
tra
ff
ic
on
node
I
f
r
om
tim
e
interval
to
p
,
w
her
e
p
=
1,2
,
.....
..
k
.
Furth
erm
or
e
,
R
p
f
or
ever
y
update
per
i
od
s
can
be w
ritt
en
a
s
(
)
=
min
=
1
…
.
(
1
−
,
)
(
7)
or
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2018
:
4467
-
4476
4472
(
)
=
min
=
1
…
,
.
(
1
−
(
1
∑
,
(
−
1
)
+
,
(
−
1
)
)
=
1
1
)
)
(
8)
Av
ai
la
ble
ba
ndwi
dth
on
eac
h
li
nk
as
pro
pose
d
in
Eq
uation
(
8
)
dif
fers
from
tradit
iona
l
avail
able
band
width
m
od
el
giv
e
n
in
E
qu
at
io
n
(
1)
.
I
n
this
m
echan
ism
,
we
con
ti
nu
ou
sly
up
date
the
avail
able
ba
ndwidt
h
base
d
on
act
ua
l
util
iz
ation
an
d
queue
le
ngth
at
ever
y
node
in
the
path
as
giv
en
in
E
qua
ti
on
(
5
)
.
T
rad
it
ion
al
routin
g
,
o
n
the
oth
e
r han
d
,
onl
y rel
ie
s
on
a
spe
ci
fic nod
e
as
giv
e
n
in
E
qu
at
i
on
(
1
)
.
2.3.
Sendin
g
R
ate
Up
d
ate
Sendin
g
rate
update
ai
m
s
to
adjust
the
sour
ce
send
in
g
rate
(
r
i
)
base
d
on
c
urren
t
net
work
conditi
on
s
.
As
m
entioned
befor
e
,
the
sou
rce
se
nd
i
ng
rat
e
ad
j
us
tm
ent
on
P
ACEC
is
ba
sed
on
the
path
rate
(
)
cal
culat
ed
by
the
SDN
c
ontr
oller.
E
ve
ry
T
C
pe
rio
d
,
the
co
ntr
oller
up
da
te
s
R
p
.
T
hen
i
n
eac
h
of
these
pe
rio
ds
,
t
he
s
ource
gets
a
ne
w
rat
e
corres
pondin
g
to
R
p
,
w
hich
is
inform
ed
by
the
con
t
ro
ll
e
r
to
the
s
ource
thr
ough
the
i
ngress
switc
h.
The
ne
w
rate
is
in
depend
e
nt
of
t
he
pr
e
vious
rate.
If
t
her
e
is
N
fl
ow
t
o
be
tra
nsm
itted
on
a
pa
th
a
nd
each f
lo
w
is given
the sam
e rate, then
R
p
on
each contr
oller
u
pdat
e is div
ided
e
qu
al
ly
. T
he
f
ollow
i
ng
e
qu
at
io
n
sh
ows
ho
w
the
sendin
g rate
(
r
i
)
is a
dju
ste
d.
(
)
=
min
=
1
…
.
(
1
−
(
∑
,
(
−
1
)
+
,
(
−
1
)
)
=
1
)
)
(
9)
Eq
uation
(9)
de
no
te
s
the
m
axim
u
m
send
ing
rate
that
can
be
pro
vid
e
d
to
tran
sm
i
t
a
fl
ow
s
o
that
the
so
urc
e
sen
ding
rate
r
i
m
eet
s
the
lim
it
≤
.
Her
e
,
we
c
on
cl
ud
e
that
to
a
dju
st
the
se
nd
i
ng
rate
at
t
he
s
ource
,
PA
CEC
us
es t
he
fo
ll
owin
g
st
eps
:
a.
The
S
DN
c
ontrolle
r
cal
culat
e
s
resou
rces
on
a
path
at
eac
h
pe
rio
d
Tc
.
T
his
res
ource
ca
lc
ulati
on
co
ve
r
s
band
width avai
la
bili
ty
w
hich
i
s the
n
c
onver
te
d
int
o path
rate
(
R
p
).
b.
Ingr
e
ss
s
witc
h
receive
s
in
f
orm
at
ion
f
ro
m
the
S
D
N
c
ontr
oller
for
eac
h
per
i
od
of
c
ontr
olli
ng
T
c
.
I
ngr
ess
switc
h passes
this in
form
at
ion
to
the
s
ource.
c.
The
s
ource
ad
justs
it
s
send
i
ng
rate
base
d
on
the
in
form
at
i
on
receive
d
f
r
om
ing
ress
sw
it
ch.
The
s
our
ce
transm
it
s
the
fl
ow
at
the
rate
,
correspo
nd
i
ng
to
the
rate
-
s
ha
rin
g
al
g
ori
thm
at
the
so
urce.
The
canno
t
exceed
,
as
the
upper
li
m
it
.
T
he
sou
rce
will
stream
the
flo
w
with
t
he
sa
m
e
rate
un
ti
l
there
is
a
c
hang
e
of info
rm
ation
rate
. If
t
her
e
ar
e N
flo
ws
t
hen
∑
1
≤
.
3.
RESU
LT
S
A
ND AN
ALYSIS
In
t
his
sect
ion,
we
re
port
the
si
m
ulati
on
res
ults
colle
ct
ed
with
ou
r
im
ple
m
entat
ion
of
P
ACEC
in
the
Mi
nin
et
Sim
ul
at
or
.
We
c
ompare
d
the
sim
ulati
on
res
ults
of
t
he
propo
sed
m
et
ho
d
w
it
h
oth
e
r
c
onge
sti
on
con
t
ro
l
m
echan
ism
s
su
ch
as
RC
P.
Her
e
we
com
par
ed
P
A
CEC
wi
th
RC
P
du
e
t
o
the
re
aso
n
that
PA
C
EC
is
a
rate
base
d
co
ng
e
sti
on
c
ontr
ol
su
c
h
as
R
CP.
W
e
al
s
o
com
par
e
PAC
EC
with
T
CP
to
see
P
ACEC
i
m
pr
ovem
ents
ov
e
rs
im
po
rtan
t
congesti
on
c
on
t
ro
l
pr
oto
c
ol
on
t
he
In
te
rn
e
t.
To
m
easur
e
the
perf
or
m
ance
of
the
co
ng
e
sti
on
con
t
ro
l
m
echan
ism
,
we
evaluated
the
t
hro
ughput,
e
ff
ic
ie
ncy,
sm
oo
th
ne
ss,
an
d
f
ai
rn
e
s
s.
As
a
ref
e
ren
ce
sce
na
rio, we
writ
e the sim
ulati
on
par
am
et
ers
in
Table
1
.
Table
1.
Param
et
ers
of
Sim
ula
ti
on
p
ara
m
et
er
v
alu
e
Bo
ttlen
eck Lin
k
C
=
1
0
0
M
b
p
s
Bu
ff
er
Size
B [
1
0
,100]
Si
m
u
latio
n
T
i
m
e
100
seco
n
d
s
E
m
u
lato
r
Minin
et
Co
n
troller
Ryu
Top
o
lo
g
y
Ab
ilen
e
Pack
et Size
1
5
0
0
By
tes
RTT
5
0
m
s
Op
erating
sy
ste
m
Ub
u
n
tu
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
Eff
ect
iv
e Ro
ute
r Assiste
d C
on
gestio
n
Co
ntr
ol
for SD
N
(
So
fi
a
N
an
i
ng H
ert
i
ana
)
4473
3.1.
Th
ro
ug
h
p
ut
Figure
4
sho
w
s
com
par
iso
n
thr
ough
pu
t
ac
hi
evem
ent
of
co
ng
e
sti
on
co
ntr
ol
sc
hem
e.
Our
sim
ulati
on
is
cond
ucted
by
enterin
g
tw
o
kinds
of
flo
w
s
consi
sti
ng
of
la
rg
e
flo
w
(f
l
ow
A
)
a
nd
sm
all
flow
(
flo
w
B)
to
th
e
netw
ork.
We
m
easur
e
the
th
rou
ghput
for
both
fl
ows.
T
he
si
m
ulati
on
res
ulted
in
the
a
ve
rag
e
of
t
hro
ughput
f
or
PA
CEC
is
48.
9
Mb
ps
,
RC
P
is
46.3
Mb
ps
,
and
TCP
is
31,
45
M
bp
s
f
or
flo
w
A
.
T
he
si
m
ul
a
ti
on
res
ults
f
or
flo
w
B
obta
in
ed
the
a
ve
rag
e
of
t
hro
ughput
;
PA
CEC
is
12.
27
M
bps,
R
CP
is
11.
58
M
bp
s
,
a
nd
TCP
is
7.8
7
Mbp
s
.
In
this
case,
PA
CEC
ou
t
perform
ed
RC
P
and
TCP.
PA
CEC
incre
ased
m
ean
thro
ug
hput
achie
vem
ent
ov
e
r
RC
P
b
y
5.7%
and TC
P
by
5
5.7
%.
Fi
gure
4.
Th
r
ough
pu
t ac
hie
ve
m
ent o
f
co
nges
ti
on
c
on
t
ro
l s
c
hem
e
3.2.
E
ff
ic
ie
nc
y
The
ef
fici
enc
y
of
the
c
ongestio
n
co
ntr
ol
m
echan
ism
can
be
deter
m
ined
thr
ough
the
po
we
r
par
am
et
ers,
i.e.
, th
e
rati
o betw
een thr
oughput
and
delay
[
24
].
=
ℎ
ℎ
(1
0
)
We
s
umm
ariz
e
the
sim
ulatio
n
res
ults
in
Table
2.
We
can
see
t
hat
P
ACEC
ha
s
the
highest
powe
r
val
ue
com
par
ed
t
o
R
CP an
d
TC
P.
T
hat
in
dicat
e
d
t
hat P
ACEC i
s
m
or
e eff
ic
ie
nt
than
RC
P
and
TCP.
Tabl
e
2
. E
ff
ic
i
ency com
par
is
on of
P
ACEC,
RC
P,
an
d TC
P
Co
n
g
estion co
n
trol
sch
e
m
e
Thro
u
g
h
p
u
t (
Mbp
s)
Delay (
m
s)
Po
wer
(α=1
)
Flo
w A
Flo
w B
Flo
w A
Flo
w B
Flo
w A
Flo
w B
PACEC
4
8
,9
1
2
,2
5
4
,2
1
2
,9
0
,9
0
,9
RCP
4
6
,3
1
1
,5
5
7
,1
1
4
,6
0
,8
0
,8
TCP
1
2
,2
7
,8
8
4
,1
1
7
,1
0
,4
0
,5
3.3.
Smoo
th
ne
ss
Sm
oo
thn
ess
is
an
esse
ntial
featur
e
of
c
onge
sti
on
co
ntr
ol.
Her
e
we
decla
red
sm
oo
th
nes
s
as
the
rati
o
of the
rate c
ha
ng
e
b
et
ween t
wo su
c
cessi
ve
update
per
i
od
s
to the p
rev
i
ou
s
r
at
e a
nd writ
te
n
as
[25
]
ℎ
=
|
−
−
1
|
−
1
(
11)
w
he
re
x
i
is
th
e
aver
a
ge
t
hro
ughput
durin
g
the
i
-
th
i
nter
val
f
or
t
he
fl
ow
(eac
h
range
is
T
c
sec).
A
flo
w
sm
oo
thn
ess
in
dex
is
def
i
ned
as
the
m
ean
throu
ghput
-
cha
nge
over
it
s
li
fetim
e
rati
o,
w
he
re
T
is
t
he
tot
al
tim
e
interval
duri
ng the sim
ulati
on
. S
m
al
le
r
s
m
oo
thn
ess
in
dices
sh
ow
sm
oo
ther
thro
ughput c
ha
ng
e
s [2
6].
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2018
:
4467
-
4476
4474
ℎ
̅
̅
̅
̅
̅
̅
=
(
∑
ℎ
=
1
)
(
12)
We
do
a
sim
ulati
on
of
10
dif
f
eren
t
fl
ows,
a
sm
a
ll
flow
ha
vi
ng
siz
es
(1
-
10
Kbps
),
m
edium
(1
0Kb
ps
-
1Mbps)
,
a
nd
la
rg
e
(
1
-
10
0Mb
ps)
within
100
seco
nds.
We
de
cl
are
t
he
th
rou
ghput
fluctu
at
io
n
with
th
e
sm
oo
thn
ess
pa
ram
et
er
as
in
(
12
)
.
Ta
ble
3
sh
ows
t
he
c
om
par
ison
of
s
m
oo
thn
ess
i
ndex
f
or
each
fl
ow.
T
he
sm
oo
thn
ess
in
dex
ca
ptures
the
tim
e
series
of
rate
c
hange
s.
A
sm
aller
sm
oo
thn
ess
in
de
x
ind
ic
at
es
s
m
oo
ther
thr
oughput c
ha
ng
e
for a
flo
w.
T
a
bl
e
3.
C
om
pa
r
i
s
on
o
f
S
m
oo
t
hn
e
s
s
(
Th
m
)
Flo
w id
Flo
w ty
p
e
PACEC
RCP
TCP
1
s
m
all
0
,05
0
,09
0
,13
2
m
e
d
iu
m
0
,03
0
,04
0
,10
3
m
e
d
iu
m
0
,02
0
,09
0
,17
4
m
e
d
iu
m
0
,02
0
,09
0
,16
5
m
e
d
iu
m
0
,02
0
,03
0
,06
6
large
0
,02
0
,08
0
,06
7
large
0
,03
0
,04
0
,10
8
large
0
,02
0
,08
0
,11
9
large
0
,02
0
,07
0
,12
10
large
0
,02
0
,06
0
,18
av
erage
0
,02
5
0
,06
7
0
,12
The
sim
ulatio
n
resu
lt
s
sh
ow
that
s
m
oo
th
nes
s
ind
ex
f
or
PAC
EC
var
ie
s
from
0.
02
to
0.0
5,
it
var
ie
s
from
0.
03
t
o
0.0
9
f
or
RC
P,
and
var
ie
s
fro
m
0.
06
to
0.1
8
f
or
TCP
.
Ba
sed
on
the
se
va
lues,
in
dicat
ing
t
hat
PA
CEC
is sm
oothe
r
c
om
par
ed wit
h
RC
P a
nd TCP
.
3.4.
Fairnes
s
Fairnes
s
is
us
e
d
to
dec
ide
w
he
ther
the
us
e
r
or
a
ppli
cat
ion
receives
a
fair
sh
are
of
syst
e
m
reso
ur
ces
.
Her
e
, we
us
e t
he
m
at
he
m
at
ical d
efi
niti
on
from
Jain and
C
hiu
[
26
]
. Fai
r
ne
ss can be
writ
te
n
as
(
x
1
,
x
2
,
…
.
.
,
x
n
)
=
(
∑
x
i
n
1
)
2
n
.
∑
x
i
2
n
1
(13)
w
he
re
x
i
is
th
e
thr
oughput
of
flo
w
i
an
d
n
is
the
su
m
of
flo
w.
We
co
nsi
der
the
m
edium
-
te
r
m
fairness
of
PA
CEC
.
T
o
e
valuate
m
edium
-
te
r
m
fairn
es
s,
we
obta
in
t
he
ave
rag
e
t
hroug
hput
of
P
ACEC
flo
ws
ov
e
r
the
entire
tim
e
of
s
i
m
ulati
on
(10
0
seco
nd
s
).
We
si
m
ulate
d
20
identic
al
flo
ws
(10
Mb
ps).
Ba
sed
on
e
qua
ti
on
(
13)
,
we
got
fairn
e
s
s
ind
e
x
for
20
identic
al
flow
s
as
sh
own
in
F
igure
5.
T
he
fa
irness
in
de
x
of
PA
CEC
is
s
m
al
le
r
than
RC
P
an
d
RC
P.
The
fair
ness
in
dex
of
PA
CEC
is
0.9
9
(close
to
on
e
)
wh
ic
h
ind
ic
at
es
that
the
through
pu
t
assignm
ent for
a co
m
peting fl
ow in
PA
CEC
is fair.
Figure
5. Fair
ne
ss in
dex of
c
onge
sti
on contr
ol sc
hem
e
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
Eff
ect
iv
e Ro
ute
r Assiste
d C
on
gestio
n
Co
ntr
ol
for SD
N
(
So
fi
a
N
an
i
ng H
ert
i
ana
)
4475
4.
CONCL
US
I
O
N
In
t
his
pap
e
r,
we
desig
ne
d
a
ne
w
R
oute
r
Assisted
C
onge
sti
on
C
on
t
ro
l
(RACC)
m
ec
han
ism
that
works
with
the
SDN
fr
am
ework.
T
his
m
echan
ism
is
designed
to
ove
rco
m
e
wea
kn
e
sses
i
n
tra
diti
on
al
Int
ern
et
netw
orks
t
hat
can
no
t
pro
vi
de
global
in
f
or
m
at
ion
net
works.
W
e
propose
a
sc
he
m
e
that
us
es
ex
plici
t
inf
or
m
at
ion
from
the
con
t
ro
ll
er
as
a
net
wor
k
poli
cy
determ
i
ner.
Ba
se
d
on
the
in
f
or
m
at
ion
from
the
co
ntr
oller,
the
se
nd
e
r
ca
n
adjust
t
he
se
nd
er
rate
acco
rd
i
ng
to
t
he
netw
ork
c
onditi
ons.
The
sen
de
r
does
not
need
to
adjus
t
the sendi
ng
rat
e increm
ental
l
y.
W
e
ha
ve
de
m
on
strat
ed
th
r
ough c
om
pu
te
r
si
m
ula
ti
on
tha
t t
he
schem
e
is able t
o
us
e
the
netw
ork
ba
ndwi
dth
m
or
e
e
ff
ic
ie
ntly
and
m
or
e
co
nsi
ste
ntly
,
and
al
so
able
to
m
ain
ta
in
fair
ness
of
a
ny
flo
w
that
re
quest
s
netw
ork
s
erv
ic
es.
For
f
uture
w
orks,
we
pla
n
to
int
egr
at
e
this
sc
hem
e
with
adm
issi
on
con
t
ro
l a
nd
b
a
ndwidt
h
al
loca
ti
on
m
echan
is
m
.
ACKN
OWLE
DGME
NTS
This
w
ork
is
par
ti
al
ly
su
pp
or
te
d
by
the
Directo
rate
of
Re
search
an
d
Com
m
un
ity
Ser
vice,
the
Gen
e
ral
Direct
or
at
e
of
Re
sea
r
ch
a
nd
Dev
el
opm
ent
Stren
gth
eni
ng
,
t
he
Mi
nistry
of
Re
sea
rch,
Tec
hnol
ogy,
an
d
Higher
Educat
i
on of
t
he
Re
pu
blic
of
I
ndones
ia
unde
r
the
r
es
earch
co
ntract
1603/K
4/KM/
2017.
REFERE
NCE
S
[1]
Van.
J.
Congestion Av
o
idanc
e
an
d
Control
.
In
SI
GCO
MM
’88.
Cal
iforn
ia
,
US
A. 1998;
18(4):
314
–
329.
[2]
John
N.
Congesti
on
cont
ro
l
in
IP
/T
CP
int
ern
et.
A
CM
SIGCO
MM
Computer
Comm
unic
ati
on
Re
v
i
ew
.
1984;
14(2)
;
11
-
17.
[3]
Manjur
K,
Mos
l
eh
M
,
Fa
iza
R.
QoS
Design
Considera
t
ion
for
Ent
erp
r
ise
and
p
rovide
r’s
Netwo
rk
at
Ingre
ss
an
d
Egre
ss
Route
r
for
VoIP
Protocol
.
Inte
rnat
ional
Journal
of
El
e
ctr
ic
al
and
Computer
Engi
ne
erin
g
(IJECE).
2016;
6(1):
235
-
241
.
[4]
Nandit
a
D,
Masa
y
oshi
K,
Rui
Z,
Nick
M.
Proce
ss
or
Sharing
Fl
ows
on
the
Inte
rnet
.
Thi
rt
eenth
Inte
rna
t
ion
a
l
W
orkshop on
Quali
t
y
of
Se
rvice
(IW
QoS
).
Pass
au,
Germ
an
y
.
200
5;
271
-
285.
[5]
Ramanja
ne
y
a
R
,
Chenn
a
R
,
P
admava
tha
m
m
a
M,
Perform
an
ce
Enha
n
ce
m
en
t
of
TCP
Frien
dl
y
Rate
Con
tr
ol
Protocol
over
W
ire
d
Networks.
Inte
rnat
ional
J
ournal
of
Elec
tric
al
and
Compu
te
r
Engi
n
ee
ring
(IJECE).
2016;
6
,
(6):2949
-
2954.
[6]
Allman
M,
p
axo
n
V Ste
vens
W
,
"TCP Conge
stio
n
Contr
ol"
,
RF
C2581,
1999
.
[7]
Sally
F.
TCP
a
nd
expl
i
ci
t
con
gesti
on
noti
f
icat
ion.
ACM
SIGC
OMM
Computer
Comm
unic
ati
on
Re
v
ie
w
.
1994
;
24(5):
8
-
23.
[8]
Juan
P,
Mounir
H
.
Enha
n
cem
ent
s
on
Route
r
-
As
sisted
Congesti
on
Contr
ol
for
W
ire
l
ess
Networks.
IE
E
E
Tr
ansacti
ons on Wirele
ss
Comm
unic
ati
ons
.
2008;
7(6):
2253
–
226
0.
[9]
Ramakrishnan
K,
flo
y
d
S,
B
lack
D.
The
Addi
ti
on
of
Expl
i
ci
t
Congesti
on
Noti
fic
a
ti
on
(
ECN)
t
o
IP.
RF
C
3168,
2001.
[10]
Allman
M,
Ja
in A
,
Sarola
h
ti P. Q
uic
k
-
Start for TCP a
nd
IP.
RF
C
4782,
2007
.
[11]
Dina
K,
Mark
H,
Charlie
R
.
C
ongesti
on
Con
trol
for
High
Bandwi
dth
-
Dela
y
Product
.
Pro
ce
ed
ings
of
Th
e
Confer
ence
on
Applic
a
ti
ons,
T
ec
hnolog
ie
s,
Ar
chi
t
ec
tur
es,
and
Protocol
s
for
Com
pute
r
Com
m
unic
at
ions.
N
e
w
York,
US
A
.
200
2;
32(4):
89
-
102.
[12]
Nandit
a
D,
Ni
ck
.
M.
W
h
y
Flow
-
Com
ple
ti
on
Ti
m
e
is
t
he
Right
M
et
ri
c
for
Congest
ion
Control
and
W
hy
Thi
s
Mea
n
s
W
e
Nee
d
New
Algorit
hm
s.
AC
M
SIGCO
MM
Computer
Comm
unic
ati
on
R
ev
i
ew
.
2006;
36(1)
:
59
-
62.
[13]
Sim
on
H,
Mich
ae
l
S,
Joche
n
K
,
Chawa
pong
S
.
Quick
-
Start
an
d
XCP
on
a
net
work
proce
ss
or:
Imple
mentat
ion
issues
and
perfo
rm
ance
ev
a
luat
i
on
.
In
Proc
.
Hig
h
Perform
anc
e
Sw
it
chi
ng
and
Ro
uti
ng.
Shanghai,
China
.
2008;
241
–
246.
[14]
Paulo
L
,
Saver
i
o
M,
Edmundo
M.
Open
Bo
x
P
rotocol
(
OBP)
.
Proce
edi
ngs
of
t
he
3rd
in
te
rn
at
io
nal
con
fer
en
ce
o
n
High
-
Perform
anc
e
Com
puti
ng
a
nd
Com
m
uni
ca
tions
.
Hous
ton,
T
exa
s.
2007
;
496
–
507.
[15]
Slim
B,
Thomas
B,
Alex
andr
e
P,
Gw
ena
ël
R
,
Jam
es
R
.
Sta
ti
sti
cal
bandwidt
h
shar
i
ng:
A
study
of
c
ongesti
on
at
fl
o
w
le
v
el
.
ACM
SIG
COM
M Com
put
er
Com
m
unic
at
i
on
Review.
New
York,
US
A.
200
1;
31,
(4):
111
–
1
22.
[16]
Monia
G,
Sohei
l
H,
Yashar
G.
R
et
hin
ki
ng
End
-
to
-
End
Congesti
on
Control
in
So
ftw
are
-
Defi
ned
N
et
works
.
in
proc
HotNets
-
IX,
Re
dm
ond,
W
ashingt
on.
2012
;
61
-
6
6.
[17]
Li
ng
y
un
L
,
Y
an
g
X,
Haif
eng
D.
OpenF
low
Contr
ol
for Coope
rat
i
ng
AQM
Sch
eme
.
Signal Proce
ss
i
ng
(ICSP
),
IEE
E
10th
Int
ern
ation
al
Conf
ere
n
ce.
B
ei
ji
ng
,
Ch
ina
.
20
10;
2560
–
2563.
[18]
Yao
H
,
Ti
ng
P
,
Li
anming
Z
.
Sof
twar
e
-
Defi
n
ed
C
ongesti
on
Contr
ol
Algorit
hm
for
IP
Networks,
H
indawi
Scientifi
c
Program
min.
2017;
2017;
1
-
9.
[19]
Masoum
eh
G,
Behz
ad
.
G.
Conge
stion
cont
rol
in
software
-
def
in
ed
data
ce
n
te
r
networ
ks
through
fl
ow
rerouting
.
i
n
proc
.
Elec
tr
ical Engi
ne
eri
ng
(
IC
EE
),
T
ehr
an
,
Ir
a
n.
2015
.
[20]
Yife
i
L,
Shuhong
Z.
SDN
-
based
TC
P
conge
st
ion
cont
rol
in
data
c
ent
er
n
et
works
.
Computing
and
Comm
unic
ati
ons
Confe
renc
e
(
IPC
CC)
,
2015
IEEE
34th
In
te
rnation
al
P
erformance
,
2015;
1
-
7.
[21]
Lu
Y,
Li
ng
Z,
Zhu
S,
Ta
ng
L.
SD
TCP:
Tow
ard
s
Datace
n
te
r
TCP
Congestion
Contro
l
wi
t
h
SD
N
for
IoT
Applic
a
ti
ons
“
,
i
n
Sensors
.
2017;
17(1):
109
.
[22]
I.
D
Ira
wat
,
Su
n
Y,
Hari
y
an
i
,
Hadi
y
oso
S
.
Link
Aggrega
ti
on
Control
Protoc
ol
on
Software
Defi
ned
Networ
k
.
Inte
rnational
Jo
urnal
of El
e
ct
ri
c
al
and
Comput
er
Engi
n
ee
ring
(IJECE). 2017; 7(5
):
2706
-
2712.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2018
:
4467
-
4476
4476
[23]
Sofia
N,
Adit
K,
Hendra
wan,
Udjia
nn
a
P.
Pat
h
A
ss
oci
ati
vity
Cen
t
ralized
Ex
pl
ic
i
t
Congesti
on
Cont
rol
(
PA
CEC)
for
SDN
.
Control,
E
le
c
troni
cs,
Ren
e
wable
Ene
rg
y
a
nd
Com
m
unic
at
i
ons
(ICCREC),
2017
Inte
rn
at
ion
al
Conf
ere
n
ce
o
n
ICCERE
C.
Yogy
ak
arta, Indone
s
ia
.
2017.
[24]
Raj
J,
Ramakris
hnan,
Dah
-
Ming
C
.
Congesti
on
Av
oidan
ce
in
Co
mputer
Net
work
s
wit
h
a
Connec
ti
onle
ss
Net
work
Layer
.
Proc
ee
di
ngs.
Com
pute
r N
et
working
S
ym
posium
,
W
ashi
ngton,
DC,
US
A
.
1987
.
[25]
Chen
K,
Nahrst
ed
K
.
Limi
tat
ion
s
of
Equation
-
ba
sed
Congesti
on
Control
in
Mob
i
le
Ad
Hoc
N
et
w
orks
.
Distributed
Com
puti
ng
S
y
st
ems
W
orkshops
,
2004.
Proce
ed
in
gs.
24th Inte
rn
at
i
onal
Conf
ere
n
ce on,
Tok
y
o,
Japa
n,
Japa
n
.
20
05.
[26]
Jain
R
,
Dah
-
Min
g
Chiu
,
W
.
Haw
e
.
A
Quan
ti
t
at
iv
e
Mea
sure
of
Fai
rne
ss
and
Discri
m
ina
ti
on
for
Res
ou
rce
Al
loc
a
ti
on
in
Share
d
S
y
st
e
m
”
,
R
ese
arch
Repor
t
TR
-
301
,
19
84.
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Sofia
Naning
gr
adua
t
ed
from
Br
awij
a
y
a
Univer
s
ity
,
Indone
sia
,
i
n
1995
and
r
ece
ive
d
th
e
M.
E.
degr
ee
from
the
Bandung
Instit
ute
of
Te
chno
lo
g
y
(IT
B)
,
Indone
sia,
in
2004,
b
oth
in
El
ectri
ca
l
Engi
ne
eri
ng,
m
aj
oring
in
T
el
e
comm
unic
at
ion
Engi
nee
r
ing.
Curre
ntly
,
she
is
pursuing
the
doct
ora
l
degr
e
e
at
the
sam
e
univ
ersity
.
Sin
ce
199
9,
she
joi
n
ed
the
Facul
t
y
of
E
lect
ric
a
l,
T
el
kom
Univer
sit
y
,
Ind
onesia
,
as
a
L
ec
tur
er.
H
er
re
sea
rch
int
er
ests
cove
r
tra
ff
ic
engi
ne
eri
ng
and
software
-
def
in
ed
net
wor
k.
Adit
Kurniawa
n
gra
duated
from
the
Bandung
Ins
ti
tute
of
T
ec
hnol
og
y
(I
TB),
Indo
nesia
,
in
1986
and
re
ce
iv
ed
the
M.E
ng.
degr
ee
from
RMIT,
Aus
tra
lia,
in
1996
and
th
e
Ph.D.
d
egr
ee
from
th
e
Univer
sit
y
of
South
Aus
tra
li
a
,
in
2003,
both
in
te
le
comm
unic
a
t
ion
engi
neering.
He
bec
am
e
a
fac
ul
t
y
m
ember
of
the
Depa
rtme
nt
of
Elec
tr
ical
Engi
ne
eri
ng,
IT
B,
in
1990.
His
r
ese
arc
h
intere
sts
are
antenna
and
wave
propa
gati
on,
ce
l
lul
a
r
comm
unic
at
ion
s
y
st
em,
and
rad
io
c
om
m
unic
at
ion.
He
is
cur
ren
tly
a
Profess
or
and
serve
s
as
the
Hea
d
of
Telec
om
m
uni
ca
t
io
n
Engi
nee
r
ing
Depa
rtment at
th
e
Bandung
Institute
of
T
ec
hnolo
g
y
.
Hendra
wan
gra
d
uat
ed
from
the
Bandung
Instit
ut
e
of
Technol
og
y
(IT
B),
Indone
si
a,
in
1985
and
rec
e
ive
d
the
M.
Sc.
degr
ee
from
Univer
sit
y
of
Es
sex,
UK
,
in
199
0and
th
e
Ph.D.
degr
ee
from
the
Univer
sit
y
of
Es
sex,
UK
,
in
199
4,
both
in
te
l
ec
o
m
m
unic
at
ion
en
gine
er
ing.
H
e
be
ca
m
e
a
fa
cult
y
m
ember
of
the
Depa
rtment
o
f
El
e
ct
ri
ca
l
Eng
in
ee
ring
,
I
TB,
in
1987.
He
is
a
m
ember
of
I
EE
E
Com
m
unic
at
ion
Socie
t
ie
s
(Co
m
Soc).
His
rese
arc
h
intere
sts
are
t
eletr
af
fic
e
ngine
er
ing
and
m
ult
imedia
n
et
w
ork.
U.
S.
Pasaribu
i
s
an
As
socia
te
Profess
or
in
Depa
rtment
of
Ma
t
hemati
cs
and
N
at
ura
l
Scie
n
ce
s,
Bandung
Instit
u
te
of
Technol
o
g
y
,
Indone
sia
.
She
gra
duated
from
the
Bandung
Instit
ute
of
Te
chno
log
y
(IT
B),
Indone
sia
,
i
n
1985and
re
ceive
d
her
Ph.D
.
from
the
Euro
pea
n
Business
Mana
gement
School,
Univ
ersi
t
y
of
W
ales,
Sw
anse
a,
UK
.
Her
rese
arc
h
i
nte
rests
include
stocha
sti
c
pro
ce
s
s,
spac
e
-
ti
m
e an
aly
s
is, and
Mark
ovia
n
m
odel
s.
Evaluation Warning : The document was created with Spire.PDF for Python.