Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
5, N
o
. 2
,
A
p
r
il
201
5, p
p
.
28
9
~
29
6
I
S
SN
: 208
8-8
7
0
8
2
89
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
The E
x
t
e
nded Di
jkstra’s-
b
ased L
oad Bal
a
ncing f
o
r OpenF
l
ow
Network
Widhi Yah
y
a
1
, Ac
hmad B
a
s
uki
2
, J
e
hn
-R
u
e
y
Ji
an
g
3
1,2
Department of
Electr
i
cal
Engin
eering
,
Un
iv
ersity
of
Brawijay
a, Malang,
Indones
i
a
1,3
Department of
Computer Scien
ce
and
Information Engin
eering
,
National Cen
t
ral Uni
v
e
r
si
ty
,
T
a
oy
ua
n City
, T
a
i
w
a
n
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Dec 26, 2014
Rev
i
sed
Feb 9, 20
15
Accepted
Feb 20, 2015
This paper
prop
oses load-balancing algor
ithm o
n
the b
a
sis of th
e Extended
Dijkstra’s shortest path algorith
m
for
Software Defined Networ
king (SDN).
The Extended D
ijkstra’s algor
ith
m consid
ers not only
th
e edge
weights, but
also the node w
e
ights to find th
e near
est server
for a requesting client.
Th
e
proposed algorithm also considers the li
nk load in order to avoid congestion
.
We use Py
r
e
tic
to implement th
e pr
oposed algo
rithm and compare it with
related ones und
er the Abilene n
e
twork
topolog
y with the Mininet emulatio
n
tool. As shown
b
y
th
e comparisons,
the proposed algorithm outp
e
rforms the
others in
term of the network
end
-
to-end latency
,
and
throughpu
t.
Keyword:
Dijk
stra’s algorith
m
Loa
d
-balanci
ng
Sh
ort
e
st
pat
h
Soft
ware
de
fi
n
e
d
net
w
or
ki
n
g
Copyright ©
201
5 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Wi
d
h
i
Y
a
h
y
a
,
Depa
rt
m
e
nt
of
El
ect
ri
cal
Engi
neeri
n
g
,
Uni
v
ersity of
Brawijaya,
Jl
. Vet
e
ra
n,
M
a
l
a
ng
6
5
1
4
5
,E
ast
Java,
I
n
do
n
e
si
a.
Em
a
il: wid
h
i
.yah
ya.ub@g
m
ai
l.co
m
1.
INTRODUCTION
The em
ergenc
e of t
h
e S
D
N t
echn
o
l
o
gy
bri
n
gs m
a
ny new net
w
o
r
k ap
pl
i
cat
i
ons
real
i
zed by
pr
o
g
ram
m
i
ng t
h
e SD
N co
nt
r
o
l
l
e
r. Ty
pi
cal e
x
am
ples include load-balanc
i
n
g
,
m
u
lti
med
i
a m
u
lticast, in
tru
s
i
o
n
det
ect
i
o
n
,
and
so on
. Loa
d
-
b
al
anci
n
g
i
s
an im
port
a
nt
conce
p
t
i
n
net
w
o
r
ki
n
g
. The
pu
r
pose
of t
h
e l
o
ad-
b
a
lan
c
ing
app
l
icatio
n
is to
d
i
strib
u
t
e th
e lo
ad
s in
to
m
u
ltip
l
e
serv
ers in
o
r
d
e
r to
g
e
t th
e
b
e
st p
e
rform
a
n
ce [1
].
Online
service
s
, s
u
ch as e
-
c
o
mmerce, e-gove
rnm
e
nt, we
b
sites, and
so
cial n
e
t
w
or
ks, of
ten use
m
u
l
tip
le
serv
ers to
ach
i
e
v
e
reliab
ility
an
d
h
i
gh
av
ailab
ility. Th
o
s
e serv
ice system
s
can
u
s
e a lo
ad
b
a
lan
c
er in
th
e fron
t
-
end
t
o
m
a
p re
que
st
s f
r
om
cl
i
e
nt
s t
o
ser
v
e
r
s.
Usi
n
g
ha
rd
wa
re l
o
a
d
bal
a
n
c
ers ca
n
be a
s
o
l
u
t
i
on,
but
i
t
s
u
f
f
ers
fr
om
t
h
e expe
nsi
v
e c
o
st
, a
n
d
t
h
e ri
gi
d
pol
i
c
y
im
posed
by
t
h
e v
e
n
d
o
r.
Th
e SD
N t
ech
n
o
l
ogy
e
n
a
b
l
e
s us
ers t
o
desig
n
t
h
eir
o
w
n
fle
x
ible a
n
d c
o
st-e
fficient
load
b
a
lan
cer th
at is su
itab
l
e fo
r th
eir
syste
m
.
Som
e
load-balancing m
e
thod
using SDN technology are
recen
tly propose
d. L
A
BER
I
O
[1] a
nd
LOBUS [2
] con
s
id
er th
e
p
a
th an
d
link
u
tiliz
atio
n
as a m
e
t
h
od
to
op
ti
m
i
z
e
th
e syste
m
t
h
rou
ghp
u
t
. The o
t
h
e
r
m
e
t
hod i
s
t
h
e servi
ce-
base
d l
o
ad-
b
al
anci
ng a
l
go
ri
t
h
m
[3]
.
Servi
ce-
base
d m
eans co
nsi
d
e
r
t
h
e fl
o
w
t
h
at
re
l
a
t
e
d
to specific se
rvices. T
h
e purpose
of
the service
-
base
d load-bala
n
cing
algorithm
is
specified
res
o
urces
(switch
e
s, lin
ks, and
serv
ers) with
p
a
rticu
l
ar serv
ices to
in
crease
reliab
ility an
d
av
ailab
ility o
f
th
e syste
m
.
Both m
e
thods
consider the
pa
th as th
e
im
portant thing in orde
r t
o
reach be
st perform
a
nce, but m
a
intaining the
pat
h
m
u
st
be clearl
y
defi
ne
d. I
n
t
h
i
s
pa
per,
w
e
pr
op
ose
d
l
o
a
d
-
b
al
anci
ng al
go
ri
t
h
m
t
h
at
i
s
ext
e
n
d
ed
by
sh
ort
e
st
pat
h
al
go
ri
t
h
m
as a m
e
t
hod t
o
cho
o
se
t
h
e
best
pat
h
.
In
20
14
, J.R. Jian
g
,
et.al, ex
ten
d
s th
e
well-k
nown Di
jkst
r
a
’s s
h
o
r
test pat
h
alg
o
rit
h
m
[4]
to co
nside
r
not
o
n
l
y
t
h
e ed
ge w
e
i
g
ht
s,
bu
t
al
so t
h
e
no
de
wei
g
ht
s f
o
r a
gra
p
h de
ri
ve
d
fr
om
t
h
e un
der
l
y
i
ng S
DN t
o
p
o
l
o
gy
.
As sho
w
n
b
y
th
e si
m
u
latio
n
resu
lts in
[4
]
,
th
e ex
tend
ed Dij
k
st
ra’s algo
rith
m
o
u
t
p
e
rfo
rm
s th
e Dij
k
stra’s
al
go
ri
t
h
m
and t
h
e n
o
n
-
wei
g
ht
ed Di
jk
st
ra’s
a
l
go
ri
t
h
m
unde
r
t
h
e A
b
i
l
e
ne
n
e
t
w
o
r
k
[5]
i
n
t
e
rm
s of en
d-t
o
-en
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
The
Ext
e
n
d
ed
Di
j
k
st
ra’s
-
bas
ed L
o
ad
B
a
l
a
n
c
i
ng f
o
r
O
p
en
Fl
ow
N
e
t
w
ork
(
W
i
dhi
Y
ahy
a)
29
0
laten
c
y. Th
is is b
ecau
s
e th
e
ex
tend
ed
Dijkstra’s al
g
o
rithm
tak
e
s edg
e
weigh
t
s as tran
sm
issio
n
d
e
lays ov
er
edge
s a
n
d
t
a
ke
s n
ode
wei
g
ht
s as
pr
ocess
d
e
l
a
y
s
ove
r
no
d
e
s, w
h
i
l
e
t
h
e
o
t
her t
w
o
al
g
o
ri
t
h
m
s
consi
d
e
r
onl
y
edge
wei
ght
s
o
r
no
wei
ght
s.
In t
h
is pa
per,
we propose a l
o
ad-balancing
algorith
m
for s
e
rve
r
s that a
r
e
located a
nd s
p
read in
wide
area S
DN
net
w
o
r
k
s
.
In t
h
e
SD
N, t
h
e
co
nt
r
o
l
l
e
r ha
s a gl
o
b
al
vi
e
w
o
f
t
h
e
net
w
or
k t
o
p
o
l
ogy
.
Thi
s
i
n
f
o
r
m
at
i
on
can be proc
ess
e
d to
get the
nearest servers
by usi
ng a
s
hortest p
a
th
al
g
o
rith
m
.
In
the pro
p
o
s
ed
algorith
m
,
we
can fi
nd t
h
e ne
arest
ser
v
ers
fr
om
t
h
e cl
i
e
nt
s
by
usi
ng t
h
e ex
tend
ed
Dijk
st
ra’s al
g
o
rith
m
th
at can
ad
ap
t
to
the
to
po
log
y
chang
e
s.
W
e
also co
n
s
i
d
er th
e link
lo
ad
fo
r
cong
estion
co
n
t
ro
l
.
We
u
s
e
Pyretic [6
] to im
p
l
e
m
en
t
th
e p
r
o
p
o
s
ed
alg
o
rith
m
an
d
c
o
m
p
are it with
th
e ro
und
-robin, the randomized load
-bala
n
cing algorithm and
t
h
e LAB
E
R
I
O
,
un
der t
h
e A
b
i
l
e
ne net
w
o
r
k [
5
]
t
opol
ogy
wi
t
h
t
h
e M
i
ni
net
[
7
]
em
ul
ati
on t
o
ol
. As
sh
o
w
n
b
y
t
h
e
com
p
ari
s
on
s, t
h
e p
r
op
ose
d
al
go
ri
t
h
m
s
out
pe
rf
orm
t
h
e ot
he
rs i
n
t
e
rm
of t
h
e net
w
o
r
k en
d-t
o
-e
nd l
a
t
e
nc
y
,
an
d
th
e thr
oug
hpu
t.
The rem
a
i
nder
of t
h
i
s
pa
pe
r i
s
orga
ni
ze
d as
fol
l
o
ws. I
n
Se
ct
i
on 2
,
we i
n
t
r
o
d
u
ce som
e
prel
im
i
n
ary
k
now
ledg
e, inclu
d
i
ng
th
e
S
DN c
o
ncept, l
o
ad-balancing
in SDN, a
n
d
th
e
ex
tend
ed
Dijk
stra’s
algorith
m
.
Sect
i
on 3
desc
ri
bes t
h
e l
o
a
d
-
b
al
anci
n
g
c
onc
ept
and t
h
e
pr
o
pos
ed al
g
o
ri
t
h
m
s
. Secti
on 4
sho
w
s t
h
e si
m
u
l
a
t
i
o
n
resul
t
s
a
n
d
obs
ervat
i
o
ns.
Fi
na
l
l
y
, t
h
i
s
pa
per
i
s
co
ncl
u
de
d
wi
t
h
Sect
i
o
n
5.
2.
PRELIMINARIES
2.
1
SD
N L
o
ad
-B
a
l
anci
ng
SD
N ad
vocat
e
s
t
h
e separat
i
o
n o
f
co
nt
r
o
l
and
dat
a
pl
anes
(or l
a
y
e
rs
), w
h
ere
un
derl
y
i
n
g
swi
t
c
h
i
n
g
har
d
ware de
vi
ces (cal
l
e
d
switch
e
s
) are con
t
ro
lled
v
i
a software en
tities (called
a
p
p
lica
tio
ns
) that ru
ns o
n
ext
e
r
n
al
, deco
upl
e
d
a
u
t
o
m
a
ted
c
ont
rol
pl
a
n
e devi
ces
(ca
l
l
e
d
con
t
ro
llers
) [8]
.
O
p
en
Flow
is one
o
f
t
h
e firs
t
ope
n pr
ot
oc
ol
s
defi
ned
bet
w
e
e
n
t
h
e
c
o
nt
r
o
l
pl
ane de
vi
ce,
t
h
e
c
ont
r
o
l
l
e
r,
a
n
d
t
h
e dat
a
pl
a
n
e devi
ce,
t
h
e
switch
,
of t
h
e SDN arc
h
itecture
[8].
An Ope
n
Fl
ow
switch
co
nsists o
f
on
e
o
r
m
o
re
flo
w ta
b
l
es
a
nd/
or
gr
o
up t
a
bl
es
, a
s
sho
w
n i
n
Fi
gu
re 2.
A
n
O
p
e
n
Fl
o
w
c
ont
rol
l
er can
u
pdat
e
,
add a
n
d del
e
t
e
fl
ow e
n
t
r
i
e
s
i
n
fl
o
w
t
a
bl
e
bot
h
reactively and
proactively. Ea
ch
flow tab
l
e in
th
e switch
con
t
ain
s
a set o
f
flow en
tries, each
of wh
ich
co
n
s
ists
of
ma
tch field
s, coun
ters,
and
set o
f
in
stru
ctio
n
s
, as show
n in
Fi
g
u
r
e
3
.
The
SD
N e
n
a
b
l
e
s t
o
desi
gn
ou
r
ow
n
p
r
ot
o
c
ol
o
n
t
o
p
o
f
SD
N s
w
i
t
c
hes.
It
i
s
a
bol
i
s
hi
n
g
ri
gi
di
t
y
o
f
n
o
wad
a
ys
n
e
twork. Th
e in
t
e
llig
en
t of
n
e
t
w
ork ad
m
i
n
i
st
rato
r
o
r
research
er can easily p
u
t
on
t
h
e
n
e
two
r
k
devices
suc
h
as loa
d
-bala
n
cing m
e
thods.N. Ha
ndig
o
l
,
et.al, [
2
] pr
opo
sed
th
e Plu
g
-n
-
S
erv
e
syste
m
im
ple
m
enting a load-balanci
ng algor
ith
m
,
called
LOBUS (LOad-Balan
c
in
g
ov
er
Un
Stru
ctur
e n
e
t
w
or
ks)
,
usi
n
g
Op
en
Fl
o
w
f
o
r
unst
r
uct
u
re
d
net
w
or
ks
.
LOB
U
S m
a
i
n
t
a
i
n
s t
h
e
net
w
or
k t
o
p
o
l
o
gy
and
l
i
n
k
st
at
us
, an
d
gree
dily choos
e
s the client-serve
r pa
ir that
yields the lo
west total re
spons
e tim
e
for
each ne
wly arriving
requ
est. M.
Ko
ern
e
r
and
O. Kao [3
] d
e
v
e
lo
p
e
d
a lo
ad
-balan
cin
g
algo
ri
th
m
fo
r h
a
nd
lin
g
m
u
ltip
le serv
ices
(cal
l
e
d LB
M
S
) by
t
h
e SD
N t
echn
o
l
o
gy
.
It
uses t
h
e Fl
ow
Vi
so
r, a
n
SDN de
vi
ce
t
o
achi
e
ve n
e
t
w
o
r
k
v
i
rtu
a
lizatio
n, to
coord
i
n
a
te m
u
l
tip
le
co
n
t
ro
llers,
eac
h
of which ha
ndles re
quests
d
es
tined
for
different
servi
ces
. I
n
2
0
1
3
,
H
.
L
o
ng
,
et
.al
,
[
1
]
pr
o
pos
ed
a loa
d
-balancing algorithm
,
nam
e
d LABERIO
(L
oAd-
B
a
l
a
ncEd R
o
u
t
i
ng w
I
t
h
O
p
e
n
Fl
o
w
)
,
t
o
m
i
ni
m
i
ze l
a
t
e
ncy and
res
p
onse
t
i
m
e
and t
o
m
a
xi
m
i
ze t
h
e ne
t
w
o
r
k
th
ro
ugh
pu
t b
y
b
e
tter u
tilizin
g
av
ailab
l
e
resources.
Fig
u
re
1
.
Th
e illu
stratio
n
of t
h
e SDN arch
it
ectu
r
e
[8
]
Fi
gu
re
2.
The
Ope
n
Fl
o
w
c
o
nt
r
o
l
l
e
r an
d t
h
e
switch
[9
]
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
5, No
. 2, A
p
ri
l
20
15
:
28
9 – 2
9
6
2
91
Fi
gu
re
3.
The
f
l
ow t
a
bl
e ent
r
y
o
f
t
h
e
O
p
e
n
Fl
ow
swi
t
c
h [
1
0]
Th
e algor
ith
m u
s
es To
R
(
T
op
of
Rack
)
Sw
itch
-
t
o
-
T
o
R
Switch
Path
s Tab
l
e
(S2SPT)
and
Lo
ad
Allo
catio
n
Tabl
e (
L
A
T
).
Ho
we
ver
,
m
a
int
a
i
n
i
n
g S
2
S
P
T
bec
o
m
e
s a
prob
lem
fo
r t
h
e LABER
I
O, if th
e
n
e
two
r
k
t
o
po
log
y
chan
ges
.
S
o
,
t
h
e LAB
E
R
I
O
i
s
n
o
t
s
u
i
t
a
bl
e i
n
wi
de
area
net
w
o
r
k
n
e
t
w
or
k t
h
at
ha
s
dy
nam
i
c t
o
p
o
l
o
gy
.
2.
2
The Ex
tende
d
Dijk
str
a
’s
Al
gori
t
hm
Gi
ve
n a wei
g
ht
ed,
di
rect
ed
gra
p
h
G
=(
V
,
E
) and a singl
e source node
s
, the classical Dijkst
ra’s
alg
o
rith
m
can
retu
rn
a sho
r
test p
a
th
fro
m
th
e source
node
s to eve
r
y othe
r node,
whe
r
e
V
is th
e set of
n
o
d
e
s
and
E
is th
e set o
f
ed
g
e
s, each
of wh
ic
h is associated
with a non-negat
i
ve
w
e
i
ght
(or
l
e
ngt
h
) [
1
1]
.
In t
h
e
o
r
i
g
in
al Dijk
st
ra’s algo
rith
m
,
n
o
d
e
s are ass
o
ciated with no weight. T
h
e pape
r [
4
]
sho
w
s h
o
w t
o
e
x
t
e
nd t
h
e
o
r
i
g
in
al algorith
m
to
co
nsid
er bo
th
t
h
e ed
ge
weigh
t
s an
d the no
d
e
wei
g
h
t
s.
Fi
gu
re 4 s
h
ow
s t
h
e ext
e
n
d
e
d
Di
j
k
st
ra
’s al
g
o
r
i
t
h
m
,
whose i
n
p
u
t
i
s
a gi
ve
n
gra
ph
G
=(
V
,
E
), the e
d
ge
weigh
t
settin
g
ew
, th
e nod
e weig
h
t
settin
g
nw
, an
d t
h
e si
ng
l
e
sou
r
ce
no
de
s
. The e
x
t
e
nde
d al
g
o
ri
t
h
m
us
es
d
[
u
]
to
sto
r
e th
e
di
s
t
ance
of t
h
e c
u
r
r
ent
s
h
ort
e
st
pat
h
fr
om
t
h
e so
urce
n
ode
s
t
o
t
h
e
dest
i
n
at
i
on
no
de
u
, a
n
d uses
p
[
u
] to st
ore t
h
e
previous
no
de prece
di
n
g
u
on
th
e curren
t
sho
r
test
p
a
th
.
In
itially,
d
[
s
]=0
,
d
[
u
]=
∞
for
u
V
,
u
s
, a
n
d
p
[
u
]
=
null fo
r
u
V
.
Exten
d
ed Dij
k
str
a
’s
Al
gori
t
hm
Inpu
t:
G
=(
V
,
E
),
ew
,
nw
,
s
Outp
ut:
d
[|
V
|],
p
[|
V
|]
1:
d
[
s
]
←
0;
d
[
u
]
←
∞
, for each
u
≠
s
,
u
V
2:
insert
u
with key
d
[
u
] into the priority queue
Q
, fo
r each
u
V
3:
whil
e
(
Q
)
4:
u
←
E
x
trac
t-Min(
Q
)
5:
for
e
ach
v
a
d
ja
ce
nt
to
u
6:
if
d
[
v
]>
d
[
u
]+
ew
[
u
,
v
]+
nw
[
u
]
then
7:
d
[
v
]
←
d
[
u
]+
ew
[
u
,
v
]+
nw
[
u
]
8:
p
[
v
]
←
d
[
u
]
Figu
re
4.
The
e
x
ten
d
ed
Di
jk
stra’s
alg
o
rithm
[4]
No
te th
at th
e ex
tend
ed
Dijkstra’s algo
rithm is
si
milar t
o
th
e o
r
i
g
in
al
Dij
k
st
ra’s algo
rith
m
.
Th
e
diffe
re
nce is that the extende
d
versi
o
n adds t
h
e node
weig
h
t
in
lin
e 6 and
lin
e 7 of th
e al
go
rith
m
.
Th
e
o
r
i
g
in
al
Dijk
stra’s alg
o
rith
m can
no
t ach
iev
e
th
e same resu
lt j
u
st
by
addi
n
g
n
o
d
e
wei
g
ht
s i
n
t
o
edge
wei
g
ht
s. T
h
i
s
i
s
because the node wei
ght s
h
ould be c
onsi
d
ere
d
only at the
outgoing edge
of an inte
rm
edia
te node on the
path.
Ad
di
n
g
n
o
d
e
wei
g
ht
s i
n
t
o
e
dge
wei
g
ht
s i
m
pli
e
s t
h
at
an ext
r
a n
o
d
e wei
ght
o
f
t
h
e
dest
i
n
at
i
on
no
de i
s
adde
d
in
to
th
e to
tal
weig
h
t
o
f
ev
ery
sh
ortest
p
a
th
,
mak
i
n
g
t
h
e al
go
rith
m
retu
rn
t
h
e
wron
g resu
l
t
.
The e
x
t
e
n
d
e
d
Di
j
k
st
ra
’s al
go
ri
t
h
m
i
s
very
u
s
eful
i
n
deri
vi
ng
t
h
e
best
r
o
ut
i
n
g
pat
h
t
o
s
e
nd
a
pac
k
et
f
r
o
m
a sp
ecif
i
c sou
r
ce
n
o
d
e
to
ano
t
h
e
r
node (
i
.e., t
h
e d
e
stin
atio
n
n
o
d
e
)
fo
r
t
h
e SD
N
env
i
ro
n
m
en
t in
w
h
ich
si
gni
fi
ca
nt
l
a
t
e
ncy
occ
u
rs
w
h
en t
h
e
packet
goe
s t
h
r
o
ug
h i
n
t
e
rm
edi
a
t
e
no
des an
d e
dge
s (o
r l
i
nks
). B
e
l
o
w
,
we
sho
w
h
o
w t
o
defi
ne t
h
e e
d
g
e
wei
g
ht
s an
d
no
de
wei
g
ht
s
so t
h
at
t
h
e e
x
t
e
nde
d Di
j
k
st
r
a
’s al
g
o
ri
t
h
m
can be
appl
i
e
d
t
o
de
ri
ve
ro
ut
i
n
g
pat
h
f
o
r s
o
m
e
speci
fi
c S
DN
en
vi
r
o
nm
ent
.
Ass
u
m
e
that we can de
rive
fro
m
th
e SDN t
o
po
log
y
a graph
G
=(
V
,
E
),
w
h
i
c
h i
s
wei
ght
ed,
di
rect
e
d
,
and
connected. For a
node
v
V
and a
n
edge
e
E
, let
Flow
(
v
) an
d
Fl
ow
(
e
)
d
e
no
te the set o
f
all th
e flows
passi
n
g
t
h
r
o
ug
h
v
and
e
, respectiv
ely, let
C
a
pab
ility
(
v
) be
the
ca
pa
b
ility
of
v
(i.e., th
e
n
u
m
b
e
r of b
its th
at
v
can proces
s pe
r second), a
n
d let
Bandw
i
d
t
h
(
e
) be t
h
e
ban
d
wi
dt
h o
f
e
(i.e., th
e nu
m
b
er o
f
b
its th
at
e
can
transm
it per second). T
h
e node wei
ght
nw
[
v
] o
f
v
is de
fined according to Eq.
(1), and
the edge wei
g
ht
ew
[
e
]
of
e
is
defi
ned
according t
o
E
q
.
(2).
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
The
Ext
e
n
d
ed
Di
j
k
st
ra’s
-
bas
ed L
o
ad
B
a
l
a
n
c
i
ng f
o
r
O
p
en
Fl
ow
N
e
t
w
ork
(
W
i
dhi
Y
ahy
a)
29
2
∑
∈
(1
)
whe
r
e B
i
t
s
(
f)
s
t
ands
f
o
r
t
h
e
n
u
m
b
er o
f
f’s
bi
t
s
pr
ocesse
d
by
n
ode
v
pe
r sec
o
n
d
.
∑
∈
(2
)
whe
r
e B
i
t
s
(
f)
s
t
ands
f
o
r
t
h
e
n
u
m
b
er o
f
f’s
bi
t
s
passi
ng
t
h
r
o
ug
h e
d
ge e
pe
r
seco
nd
.
Not
e
t
h
at
we c
a
n easi
l
y
o
b
t
a
i
n
t
h
e
n
u
m
b
er
of
a fl
ow
’s
bi
t
s
p
r
oce
ssed
by
a n
o
d
e
or
pa
s
s
i
n
g
t
h
r
o
ug
h
an e
dge
wi
t
h
t
h
e
hel
p
of
t
h
e
“co
unt
e
r
s fi
el
d”
of t
h
e
Ope
n
Fl
o
w
s
w
i
t
c
he
s’ fl
ow
t
a
bl
es.
Al
so
n
o
t
e
t
h
a
t
t
h
e
num
erat
ors i
n
Eq.
(1
) a
n
d
Eq
. (
2
) a
r
e
of t
h
e
uni
t
of “
b
i
t
s
”,
and
t
h
e
den
o
m
i
n
at
ors are
o
f
t
h
e
uni
t
of “
b
i
t
s
pe
r
seco
nd”
. T
h
ere
f
o
r
e, t
h
e
no
de
wei
g
ht
n
w
[v]
a
n
d
t
h
e e
d
ge
we
i
ght
e
w
[e]
a
r
e
of
t
h
e
u
n
i
t
o
f
“
s
eco
nds”
.
Whe
n
we
accum
u
late all the node wei
g
hts and a
ll the edge
weights a
l
ong a path, we
can obtain the en
d-to-end latency
fr
om
one e
n
d
t
o
t
h
e
ot
her
en
d
o
f
t
h
e
pat
h
.
3.
PROP
OSE
D
LOAD
-BAL
A
NCI
NG
AL
G
O
RITH
M
The loa
d
balancer is place
d i
n
the
front-e
n
d of
t
h
e online
services system
s
to
m
a
p
th
e requ
est fro
m
t
h
e cl
i
e
nt
s t
o
the ser
v
ers
.
In t
h
e Li
nu
x vi
rt
u
a
l
server (L
VS
) use Net
w
o
r
k
Ad
dr
ess Tra
n
sl
at
i
on(
N
A
T) t
o
di
rect
traffic from
the Internet to a vari
able number
of serve
r
s
on the sec
o
nd
layer, wh
ich in
tu
rn
p
r
o
v
i
d
e
th
e
necessa
ry servi
ces. Se
rvice
re
que
sts arriving at the L
V
S cl
ust
e
r a
r
e a
d
dre
ssed t
o
a
vi
rt
u
a
l
IP a
d
dress
o
r
V
I
P
.
VIP is can
be a p
u
b
lic IP ad
dress th
at asso
ciates with
a fu
lly-qu
a
lified
domain
n
a
m
e
an
d
wh
ich
is assig
n
e
d
to
m
u
l
tip
le serv
ers [1
2
]
.
Fi
gu
re
5.
Exa
m
pl
e Server
L
o
ad-balancing Setup
[13]
Fi
gu
re 5 e
x
pl
ai
ns t
h
e ser
v
er
l
o
ad-bala
n
cing
architecture by
using VI
P. I
n
t
h
i
s
pa
per
,
we use
VI
P
address
as IP
public that
will be accesse
d
from
the clie
nts. Figure
5 is an e
x
am
ple of a si
ngle
point loa
d
balance
r
arc
h
itecture.
In t
h
is pape
r, the load-bala
n
cing
algo
rith
m
was i
m
p
l
em
en
ted
o
n
t
h
e Ab
ilen
e
topo
log
y
that eve
r
y swit
ches i
n
the
topo
logy act as a
load bala
ncer.
Usi
n
g nai
v
e al
go
ri
t
h
m
s
, such
as t
h
e ro
u
nd
r
obi
n al
g
o
ri
t
h
m
and t
h
e ra
nd
o
m
i
zed al
gori
t
h
m
,
i
n
wi
de-
area-n
e
two
r
k
lo
ad-b
alan
cing h
a
s th
e po
ssi
b
ility to
fo
rward
a requ
est to
th
e fart
h
e
st
serv
er. Th
is is n
o
t
efficien
t b
ecause th
e requ
est an
d
th
e rep
lied d
a
ta w
ill g
o
acro
ss th
e
n
e
two
r
k
and
con
s
ume a
lo
t o
f
b
a
nd
wi
d
t
h
of
t
h
e
w
h
ol
e
net
w
or
k.
I
n
S
D
N
,
t
h
e c
ont
r
o
l
l
e
r
has t
h
e
g
l
obal
i
n
f
o
rm
at
ion
of
t
h
e
w
h
o
l
e net
w
or
k,
a
n
d ca
n
deci
de t
o
f
o
r
w
ard t
h
e re
que
st
t
o
t
h
e nea
r
est
serve
r
by
find
in
g
t
h
e sh
ortest p
a
th
with
th
e
ex
tend
ed
Dijk
stra’s
alg
o
rith
m
fro
m th
e clien
t
to
a serv
er,
wh
ere th
e shortest
p
a
th
m
ean
s th
e p
a
th
with
th
e sm
allest su
mm
a
t
i
o
n
o
f
no
de
wei
g
ht
s a
n
d
ed
ge
wei
g
ht
s.
Fi
gu
re
6 s
h
ow
s t
h
e
pr
o
p
o
s
ed
l
o
ad
-
b
alancing algorithm
.
The
basic idea
is
to forward
ea
ch request t
o
th
e n
e
arest serv
er with th
e
li
n
k
load
(u
tilizatio
n
o
f
th
e link
b
e
tween
th
e
serv
er and
th
e
switch
)
lower
th
an
a
pre
-
s
p
eci
fi
ed t
h
res
h
ol
d
θ
.
Howev
e
r, if all the serv
ers h
a
v
e
lin
k
lo
ad
s larger th
an
t
h
e thresho
l
d
,
th
e al
go
rith
m
still choose
s
the nea
r
est serve
r
.
In this
way,
we ca
n
pre
v
e
n
t conge
stion on
the serve
r
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
5, No
. 2, A
p
ri
l
20
15
:
28
9 – 2
9
6
2
93
Prop
osed
L
o
a
d
-B
al
anci
n
g
A
l
gori
t
hm
Inpu
t:
sw
src
,S
dst
,
Outp
ut
:
s
,
s
S
dst
P
←
eD
ij
ks
tra
(
sw
src
,
S
dst
)
for
every
p
i
P
if
p
i
.server.kl >
then
move
p
i
fro
m
P
to
Q
if
P=
th
en
s
←
mi
n
(
P
).
se
rve
r
else
s
←
mi
n
(
Q
).
se
r
v
e
r
re
tur
n
s
Fi
gu
re
6.
Pr
o
p
o
se
d L
o
ad
-balancing Algorithm
We assum
e
server
s
i
is attach
ed
to
switch
sw
i
and a switch is attached with at
m
o
st one
server. Later
on
, we
use
s
i
a
nd
sw
i
exc
h
a
n
gabely for c
o
nvenience. Gi
ven the source s
w
itch
sw
src
to
wh
i
c
h
th
e
requ
est
clien
t
is attach
ed
, th
e set
S
dst
of
ser
v
ers, a
n
d a
pr
es
pecified threshold
, t
h
e
p
r
o
p
o
s
ed
algo
rith
m
will return th
e b
e
st
serve
r
fo
r
l
o
ad
-bala
n
cin
g
.
Th
e lin
k
lo
ad
kl
i
of ser
v
e
s
i
(t
h
e
u
tilizatio
n
of th
e lin
k
<
s
i
,
sw
i
> between the serve
r
s
i
an
d
th
e switch
sw
i
)
is d
e
fin
e
d
as
fo
llo
ws:
,
,
(3
)
Since the
propose
d algorithm
is ba
se
d
on t
h
e e
x
tended
Dijkstra
’s al
gori
thm
[4], we al
so take
the
sam
e
m
echanism
s
to obtain t
h
e node wei
ghts an
d
th
e ed
ge weigh
t
s. No
t
e
th
at
eDijkstra
(
sw
src
, S
dst
) will u
s
e
th
e ex
tend
ed
Dijk
stra’s alg
o
rith
m
to
retu
rn a set
P
of shortest paths from
the source
switch
sw
src
,
to
ev
ery
serv
er in
th
e
serv
er set th
e
S
ds
t
,.
Also
no
te th
at
p
i
.
se
rver
st
ands for the
se
rve
r
as
s
o
ciated with the
path
p
i
, a
nd
hence
p
i
.
server
.
kl
stan
ds fo
r th
e link
lo
ad
o
f
th
e
serve
r
ass
o
ciated
with the path
p
i
.
F
u
rt
herm
ore, t
h
e f
unct
i
o
n
min
(
P
) will ret
u
rn t
h
e s
h
ortes
t
one
am
ong
all shortest
paths
in
P
.
4.
SIMULATION
4.
1.
Si
mul
a
ti
on
Se
tti
n
g
Accord
ing
to
th
e Ab
ilen
e
core to
po
log
y
, we set u
p
an
Op
en
Fl
o
w
co
ntro
ller and
11 Op
en
Flo
w
swi
t
c
hes as
no
des, eac
h o
f
whi
c
h was l
i
n
ked t
o
t
h
e c
ontro
ller log
i
cally, as sh
own
in
Figu
re
7
.
For lo
ad-
balancing testing, we ass
u
m
e
d there are two web servers
that placed and sprea
d
in two
diffe
re
nt locations in
the Abilene
net
w
ork that
will be accesse
d
from
c
lients.
Fig
u
r
e
7
.
Topolo
g
y
U
s
ed
in Si
m
u
latio
n
Oi
C
ont
r
o
l
l
e
r
O
p
e
n
Vswitc
h
Serve
r
Clien
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
The
Ext
e
n
d
ed
Di
j
k
st
ra’s
-
bas
ed L
o
ad
B
a
l
a
n
c
i
ng f
o
r
O
p
en
Fl
ow
N
e
t
w
ork
(
W
i
dhi
Y
ahy
a)
29
4
The POX was
used as OpenFlow c
ontrolle
r. Th
e loa
d
-ba
l
ancing algorithm
was im
p
l
e
m
ented usi
ng
Pyretic. The
VIP a
d
dress was use
d
as IP
p
ublic that will be
accessed
from
the Client.
Tabl
e 1. Si
m
u
lat
i
on
a
n
d
Set
t
i
ngs
Para
m
e
ter
Setting
Bandwidth on edg
e
s
1Gbps
Nu
m
b
er
of ser
v
er
2
Nu
m
b
er
of switches
11
Nu
m
b
er
of edges
25
Contr
o
ller
POX 2.
0 suppor
ting Py
r
e
tic
Openflow Switch
OpenVSwitch 1.
0
T
e
sting tool
I
p
er
f,
Netper
f
T
e
sting tim
e
30second
I
n
th
e
F
i
gu
r
e
7,
th
er
e
a
r
e 2 s
e
r
v
er
s
and
1
2
c
lie
n
t
s
in
t
h
e
Abilen
e
to
po
l
o
g
y
. In
th
is
sim
u
lat
i
o
n
testing,
we ge
nerat
e
d
Tran
sm
i
ssi
on C
ont
r
o
l
Pr
ot
oc
ol
(TC
P
) dat
a
s
t
ream fro
m
th
e clien
t
s to
th
e serv
ers using
Iperf. To
ex
tend
m
o
re in
fo
rm
atio
n
we also
u
s
ed
th
e Netp
erf to
g
e
n
e
rate requ
est fro
m
cl
ien
t
s. Fo
r ev
ery testing
,
we
defi
ned t
h
e n
u
m
ber of cl
i
e
nt
s are 4,
8, a
nd
1
2
cl
i
e
nt
s. T
h
i
s
sim
u
l
a
t
i
on ran
on
AM
D
Phe
n
om
(tm
)
965
0
Qua
d
-
Co
re Pro
cessor and
8GB
o
f
RAM. To prov
e
th
e algo
ri
t
h
m
’
s reliab
ility, we co
m
p
ared it with
th
e
ro
und
ro
b
i
n
and
ra
n
dom
i
z
ed al
g
o
r
i
t
h
m
s
whi
c
h
ha
ve
no
c
onsi
d
er
t
h
e
sh
o
r
t
e
st
pat
h
.
4.
2.
Si
mul
a
ti
on
Re
sul
t
In
t
h
is exp
e
rimen
t
th
e req
u
est can
h
a
pp
en
in
ev
ery switch
in
th
e t
o
po
log
y
th
at describ
e
in the
sim
u
l
a
t
i
on set
t
i
ngs
. T
h
e
pr
o
p
o
se
d al
g
o
ri
t
h
m
ch
oo
ses t
h
e
b
e
st
pat
h
(
n
eare
s
t
ser
v
er
) f
o
r a
re
quest
i
n
g
cl
i
e
nt
by
u
s
ing
th
e ex
tend
ed
Dijk
stra’s
alg
o
rith
m
.
It is
m
o
re con
v
e
n
i
en
t th
an
m
a
in
tain
in
g
t
h
e S2SPT and
th
e LAT lik
e
i
n
t
h
e
LAB
E
R
I
O.
T
h
e L
A
B
E
R
I
O
was
des
i
gne
d
fo
r a
dat
a
cen
ter th
at
used
term
o
f
To
p-o
f
-Rack
switch
to
main
tain
th
e path
s. Si
n
ce th
e
d
i
fferen
t
to
po
l
o
g
y
, in
th
is
exp
e
rim
e
n
t
we use all switch
to m
a
in
tain
th
e path
for
th
e LA
BERIO
.
A
s
show
n in
t
h
e Fi
g
u
r
e
8
,
the pr
opo
se
d algo
rith
m
o
u
t
p
e
rfo
rm
s th
e LABERIO in
t
h
e term
o
f
latency. It is
quite sim
ilar because
bot
h al
gorithm
s
consi
d
er t
h
e link c
a
pacity, but the superi
ority of the
propose
d
al
gorith
m
isalsoconsider
t
h
e node
capacity. W
e
also
c
o
m
p
are
th
e
p
r
op
o
s
ed
al
g
o
rith
m
with
n
a
iv
e
algorithm
,
such as round-robin an
d ra
nd
o
m
i
zed al
gori
t
h
m
t
h
at
do not
conside
r
the nearest serve
r
. The
si
m
u
latio
n
resu
lts sho
w
t
h
at th
e p
r
opo
sed
alg
o
rith
m
is b
e
tter th
an
th
e t
w
o
n
a
iv
e al
g
o
rith
m
s
in
th
e t
e
rm
of
end
-
t
o
-en
d
l
a
t
e
ncy
,
as s
h
o
w
n
i
n
Fi
g
u
re
8. T
h
e net
w
o
r
k en
d-t
o
-e
nd l
a
t
e
n
c
y
was m
easured by
usi
n
g t
h
e
pi
n
g
tool to se
nd 30
packets
whose pac
k
et size
is 65507
b
y
tes fro
m
clien
t
s to
th
e
serv
ers
for
3
0
seco
nd
s. Th
e
superiority is because the proposed
al
gorithm
considers
the shortest
pa
ths
(nea
rest
serve
r) a
nd a
l
so the
co
ng
estion
con
t
ro
l. On
th
e
co
n
t
rary, it is p
o
s
sib
l
e fo
r t
h
e n
a
iv
e algorith
m
s
to
fo
rward
req
u
ests to
th
e far
serve
r
t
o
g
o
t
h
ro
u
gh s
o
m
e
swi
t
c
hes. I
n
t
h
e
real
net
w
o
r
k,
a hi
gh
per
f
o
r
m
a
nce IP r
out
ers an
d swi
t
c
h
e
s ad
d
app
r
oxi
m
a
t
e
ly
20
0 m
i
croseco
nds
o
f
l
a
t
e
ncy
t
o
t
h
e l
i
n
k
d
u
e
to
packet
proc
essing. It m
eans, if t
h
e re
ques
t are
d
e
flected to
t
h
e fart
h
e
st serv
er to go
t
h
ro
ugh a lo
t
o
f
sw
itches, th
e laten
c
y
w
ill in
crease sig
n
i
fican
tly.
Th
e thro
ugh
pu
t m
easu
r
em
e
n
t and
co
m
p
arison
was co
nd
u
c
ted
to
ev
al
u
a
te th
e cap
a
b
ility o
f
th
e
propose
d
algorith
m
.
Throughput is the
rate of s
u
ccess
f
ul
message delive
r
y ove
r a c
o
mm
unication channel.
In
Fig
u
r
e
9
,
sh
ow
s th
e p
r
op
osed
alg
o
r
ith
m h
a
s h
i
gh
er
th
rou
ghp
u
t
th
an
th
e r
oun
d
ro
b
i
n
,
th
e r
a
ndomized
al
go
ri
t
h
m
,
and t
h
e LAB
E
R
I
O. Th
e ro
u
nd
ro
bi
n a
nd ra
n
d
o
m
i
zed al
gori
t
h
m
m
a
y defl
ect
request
t
o
t
h
e fa
r
serve
r
and e
v
e
n
the link is c
o
nge
sted.
It causes the thro
u
ghp
u
t
is lo
wer than
th
e pro
p
o
s
ed
algo
rith
m
.
Co
n
s
id
er
th
e nod
e cap
a
city d
e
r
i
v
e
t
h
e
pr
opo
sed h
a
s
b
e
tter
th
ro
ugh
pu
t
th
en th
e
LA
B
E
RI
O.
As s
h
ow
n i
n
t
h
e Fi
g
u
re
1
0
,
w
e
use
st
an
dar
d
devi
at
i
o
n t
o
m
easure
t
h
e se
r
v
er l
o
a
d
vari
at
i
o
n.
St
an
dar
d
devi
at
i
o
n m
easures
t
h
e am
ou
nt
o
f
vari
at
i
o
n
or
di
s
p
ersi
on
f
r
om
t
h
e ave
r
ag
e. T
h
e l
o
a
d
i
n
f
o
rm
at
i
on i
s
p
r
ovi
de
d
b
y
th
e Ip
erf serv
er p
r
og
ram
.
In
th
is exp
e
ri
men
t
, th
e p
r
opo
sed
alg
o
rith
m sh
ows go
od
resu
lt (sm
a
l
l
est
)
in
th
e
s
e
r
v
er
’s
lo
ad
va
r
i
a
tio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
5, No
. 2, A
p
ri
l
20
15
:
28
9 – 2
9
6
2
95
Fig
u
r
e
8
.
End-
to
-
E
nd
Laten
c
y
Fi
gu
re 9.
Th
r
o
ug
h
put
Fi
gu
re 1
0
. St
an
dar
d
De
vi
at
i
o
n
Loa
d
of
Se
rve
r
s
5.
CO
NCL
USI
O
N
This
pa
per
proposes
a l
o
ad-balancing algorithm
th
at
t
a
kes ad
va
nt
age
of
t
h
e e
x
t
e
nde
d
Di
j
k
st
ra
’s
sh
ortest p
a
t
h
alg
o
rith
m
.
Th
e ex
tend
ed
Dijk
stra’s sh
orte
st path algorithm
was used t
o
fi
nd t
h
e nea
r
est
serve
r
for a requ
esting
clien
t
. Th
e ex
tend
ed
Dijk
stra’s algo
rith
m
c
o
n
sid
e
rs n
o
t
o
n
l
y th
e ed
ge weigh
t
s b
u
t
also
th
e
no
de
wei
ght
s
fo
r a
g
r
a
p
h
de
ri
ve
d
fr
om
t
h
e u
n
d
e
rl
y
i
ng
S
D
N
t
o
pol
o
g
y
.
We
use
Py
ret
i
c
t
o
i
m
pl
em
ent
t
h
e
pr
o
pose
d
l
o
ad
-
b
al
anci
n
g
al
go
ri
t
h
m
unde
r t
h
e Abi
l
e
ne net
w
or
k t
o
p
o
l
o
gy
w
i
t
h
t
h
e M
i
ni
net
em
ul
ati
on t
o
ol
. The
sim
u
l
a
t
i
on res
u
l
t
s
sh
ow t
h
at
t
h
e p
r
o
p
o
se
d
l
o
ad
-bal
a
n
ci
n
g
al
g
o
ri
t
h
m
out
pe
rf
orm
s
ot
hers i
n
t
e
rm
s of t
h
e
n
e
two
r
k
end
-
to-
e
nd
laten
c
y, an
d thr
oug
hpu
t.
REFERE
NC
ES
[1]
H. Long,
Y
.
Sh
en*, M. Guo,
and
F
.
T
a
ng. “LABERIO: D
y
namic load-b
al
an
ced routing
in O
p
enFlow-enabled
networks”.
IEEE 27th In
ternation
a
l Confer
ence o
n
Adv
anced In
fo
rmation Networking and
App
l
ications
. 2013.
[2]
N. Handigo
l, S.
Seethar
a
man, M. Flaj
slik
, N
.
McKeown, and
R
.
J
ohari. “Pl
ug-n-Serve: Lo
ad-balancing w
e
b
traf
fic
using Open Flo
w
”.
Demo at
ACM SIGCOMM
, Aug.
2009.
[3]
M. Koerner
and
O. Kao. “
M
ultip
le Serv
i
c
e
Load-
B
alan
cing wi
th
Open Flow”.
I
E
EE 13th
International Confer
ence
on High Performance Switching
and Rou
ting
. 20
12.
[4]
J
.
R. J
i
ang
,
H.W
.
Huang, J
.
H.
Li
ao,
and S
.
Y
.
Ch
en. "
Extend
ing
Dijkstra’
s
Shortest Path
Algorithm for Softwar
e
Defined
Network
i
ng
". in
Proc. of
the 16
th
APNOMS. 2014.
[5]
Abilene Networ
k, http://en
.
wikipedia.or
g/wiki/A
bilen
e
_Network-
#
cite_not
e-
line-
1, last
accessed
on March
4, 201
4.
[6]
J. Reich, C. Mon
s
anto, N
.
Foster
, J. R
e
xford, an
d
D.
W
a
lker
, “Modular SDN Progr
amming with P
y
retic”.
T
echn
i
cal
Repr
ot of USENI
X
, ava
ilab
l
e
ath
t
t
p
://
www
.us
e
nix.o
r
g
, 2013.
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
2.1
48
1
2
Millise
c
onds
Number
of
Clie
nts
Proposed
Round
Robin
Randomized
LABERIO
570
620
670
720
48
1
2
Mbps
Number
of
Clie
nts
Proposed
Round
Robin
Randomized
0
0.5
1
1.5
2
2.5
48
1
2
Number
of
Clie
nts
Proposed
Round
Robin
Randomized
LABERIO
Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE
ISS
N
:
2088-8708
The
Ext
e
n
d
ed
Di
j
k
st
ra’s
-
bas
ed L
o
ad
B
a
l
a
n
c
i
ng f
o
r
O
p
en
Fl
ow
N
e
t
w
ork
(
W
i
dhi
Y
ahy
a)
29
6
[7]
Mininet W
e
bsite,
http
://mininet.o
r
g
/, last
accessed
on May
2014.
[8]
Open Network F
oundation (ONF)
W
e
bsite (SDN white p
a
pe
r)
, ht
t
p
s://www
.opennetworking.or
g/sd
n-resources/sdn-
definition, last accessed on
Janu
ar
y
2014.
[9]
Open Networkin
g
Foundation
.
“
O
penFlow Switch
Specif
i
cation
v
e
rsion 1.4
.
0”. October 14
, 2013
.
[10]
N. McKeown,
et. al. “Open Flow: Enab
ling I
nnovation
in C
a
mpus Networks”.
ACM S
I
GCOMM Computer
Communication
. 2008.
[11]
E. Dijkstra, “A
note on
two pro
b
le
ms in conn
ex
ion with gr
aphs”.
Nume
risc
he mathe
m
atik
. vo
l.
1, no.1
,
1959
, p
p
.
269-271.
[12]
Red Hat, In
c. “Linux
V
i
rtual Ser
v
er
Administration: RHEL5
:
Lin
ux
V
i
rtual Server (L
VS)”. 2007
.Doc
[13]
Wi
k
i
Cisco,
http://docwiki.cisco.com
/wiki/C
isco_ACE_4700_Series_Applian
ce_Qu
ick_Start_
G
uide,_Releas
e_
A3(1.0)_--
_Configuring_Server_Lo
ad_Balancing
,
last
accessed on june 201
4.
BIOGRAP
HI
ES OF
AUTH
ORS
W
i
dhi Yah
y
a r
ece
ived the M
a
s
t
er S
c
ienc
e de
gree in Depart
m
e
nt of Com
p
uter S
c
ien
ce and
Information Eng
i
neer
ing, N
a
tion
a
l Central Univer
sity
,
Taiwan
in
2014 as an
Inter
n
ation
a
l Dual
Degree Master student betw
een
University
of Br
awijay
a
and National Centr
a
l University
. He got
Bachelor degr
ee in Dep
a
rtment
of Informatics
E
ngineer
ing, University
of
Brawijay
a, Indon
esia.
His research interest area is in th
e computer sc
ience and informatio
n technolog
y
ar
eas, especially
in network progr
amming.
Achmad Basuki is a sen
i
or lecturer in
Depar
t
me
nt of El
ec
tri
cal
Engin
eering
,
Univers
i
t
y
of
Brawijay
a
, Indo
nesia. He is ex
pert in computer
networking re
s
earch ar
ea
, es
peci
all
y
in IP
m
u
lticast
,
P2P, CDN, data cen
ter
,
and networking
. He got Ph. D. degree from KEIO University
,
Japan. He presently
work in PTIK, University
of Brawijay
a
, I
ndonesia as a C
h
airman of the
P
T
IK
U
B
.
Jehn-Ruey
Jiang
received his Ph. D. degree in
C
o
mputer Science in 1995 from Nation
a
l Tsing-
Hua Univers
i
t
y
,
Ta
iwan,
R.O.C
.
He
join
ed Chu
ng-Yuan Chris
t
i
a
n Univers
i
t
y
as
an As
s
o
cia
t
e
Professor in 199
5. He joined Hsuan-Chuang Univ
ersity
in 1998
and became a full Professor in
2004. He is curr
ently
with the D
e
partment of
Co
mputer Science
and In
formation
Engineering
,
National Centr
a
l University
, an
d co-leads
the
Adaptive Comp
uting and Networking (ACN)
Laborator
y
,
which aims at deve
l
oping adapt
i
ve
m
echanis
m
s
for
coll
aborat
ive co
m
puting entit
ies
to make proper adjustments according to th
eir
current understandings about the computing
environments or
resources, in o
r
d
e
r
to
eff
i
ci
entl
y
perform
given
ta
s
k
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.