Indonesi
an
Journa
l
of El
ect
ri
cal Enginee
r
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
12
,
No.
3
,
Decem
ber
201
8
, p
p.
1
17
9
~
11
86
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
2
.i
3
.pp
1179
-
1186
1179
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
An Effi
cient F
ramework
to
Imp
rove QoS
of CSP us
ing
Enh
anced Min
imal R
esou
rce Opti
mizati
on
based S
ch
eduli
ng Algorit
hm
Ravi
Mahade
van
1
,
Neelame
gam Anb
az
hagan
2
1
Dept.
of
Com
pute
r
Sc
ie
nc
e and Engi
ne
eri
ng,
Al
aga
ppa
Univer
si
t
y
,
Kar
ai
kudi
,
Tam
il
nadu,
Ind
ia
2
Alaga
ppa
Univ
ersity
,
Kara
ikudi,
T
amilna
du
,
Ind
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
ul
30, 2
018
Re
vised
Sep
14, 2
018
Accepte
d
O
ct
21, 201
8
Online
Now
aday
s,
th
e
enterpri
ses
&
indi
v
idu
al
s
are
con
tri
b
uti
ng
their
workloads
on
cloud
servic
e
pro
vide
rs
which
ar
e
going
to
inc
r
e
ase
on
dail
y
basis.
Th
ere
ar
e
la
rg
e
amount
CS
P
are
ava
i
lable
to
off
er
v
irtualize
d
and
d
y
nami
c resou
rc
e on
p
a
y
and
use
basis.
How
eve
r
,
the
r
e ar
e
al
m
ost
CS
P
fai
led
to
m
ai
nt
ai
n
qua
lit
y
of
servi
ce
(QO
S)
and
m
ini
m
al
r
esourc
e
opti
m
iza
ti
on.
Som
e
of
the
ex
isti
ng
a
pproa
che
s
are
hi
ghl
y
ded
ic
a
te
d
o
n
sche
dul
ing
po
l
icy
bu
t,
it
does
not
consid
ere
d
re
li
ab
le
se
rvic
es
w
it
h
opt
i
m
iz
ed
QO
S.
To
offe
r
best
soluti
on
of
abo
ve
probl
em,
th
e
fra
m
ework
pr
oposes
Enha
nc
e
d
Minim
al
Resourc
e
Optimi
za
t
ion
base
d
Sch
edul
ing
Algor
it
h
m
to
m
ini
m
iz
e
th
e
resourc
es
and m
ai
nt
ai
n
the
QO
S.
Th
e m
et
h
od
avoi
ds
d
ela
y
i
n
Reque
st
-
R
esponse m
ode
l
in
cl
oud
envi
ron
m
ent
.
To
avoi
d
o
ver
loa
d
fo
r
resou
rce
al
lo
ca
t
ion,
th
e
proposed
design
utili
ze
d
opti
m
iz
ed
sche
d
uli
ng
pol
icy
.
Pr
oposed
m
ec
hanis
m
s
uti
li
zed
opti
m
iz
ed
servi
ce
broke
r
ing
po
li
c
y
to
r
educ
e
t
he
d
el
a
y
r
esponse
in
cl
ou
d
envi
ronm
ent
.
Th
e
fra
m
ework
al
s
o
hel
p
c
loud
user
to
pre
fer
b
est
CS
P
ac
cor
ding
to
the
i
r
prior
s
erv
ices.
Th
e
m
et
hod
offe
rs
risi
ng
tre
nd
of
r
esourc
e
base
d
struct
ure
to
red
u
ce t
he
pl
acem
ent c
hurn
ex
te
nsive
l
y
.
Propos
ed
s
y
s
t
em ut
iliz
ed
eff
icient
sche
dul
ing
polic
y
to
trans
m
it
dat
a
r
eques
t
to
CS
P
with
m
ini
m
al
da
t
a
proc
essing
ti
m
e
.
The
ent
ir
e
ut
il
i
z
at
ion is t
o
improve
th
e
QO
S of
cloud servi
c
e
provide
r
in
th
e
f
ea
tur
es
of
m
ulti
-
dimensional
r
esourc
e.
Based
on
e
xper
imental
eva
lu
at
ions,
pro
posed
t
ec
hniqu
e
improves
the
CP
T
(Com
putation
Proce
ss
ing
Ti
m
e)
301.
72
m
il
li
sec
onds,
B
U
(Bandwidt
h
Util
izati
on
)
20
Mbps
,
CP
UU
(CPU
Util
izat
io
n)
5%
&
MRU
(
Mem
ory
R
esourc
e
Utilization)
3
%
on
give
n
input
p
ara
m
eters c
om
par
e
tha
n
e
xisti
ng
m
et
hodo
l
og
y
.
Ke
yw
or
ds:
Cl
oud
c
om
pu
ti
ng
Cl
oud
s
e
rv
ic
e
p
r
ovider
m
e
m
or
y
Re
so
urce
a
ll
oc
at
ion
v
irtual
iz
at
ion
Re
so
urce
o
ptim
iz
at
ion
CPU
util
iz
at
ion
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
:
Ra
vi Maha
devan,
Dep
a
rtm
ent
of
Com
pu
te
r
Scie
nce a
nd E
ng
i
ne
erin
g
,
Alaga
pp
a
Univ
ersit
y
,
Kar
ai
kudi
, Ta
m
il
N
adu
.
Em
a
il
:
rav
i
m
a
had
e
va
n.phd
@
gm
ail.co
m
.
1.
INTROD
U
CTION
Nowa
days,
N
owadays
,
the
e
nter
pr
ise
s
an
d
in
div
id
uals
a
r
e
co
ntributi
ng
their
w
orklo
a
ds
t
o
cl
oud
serv
ic
e
pro
vide
rs
(C
SP)
ha
ve
bee
n
rap
i
dly
increase
d
on
da
il
y
basis.
CSP
structu
re
a
hu
ge
pool
of
a
bs
tr
act
ed,
virtu
al
iz
ed
, a
nd
dynam
ic
al
l
y
scal
able
res
our
ces
f
or
use
rs
,
pa
y
an
d
us
e
basi
s. T
he
res
ource
s
are
pa
rtit
ion
e
d i
nt
o
three
kinds
of
serv
ic
es:
I
nfra
structu
re
as
a
Ser
vice
(I
aa
S),
Plat
fo
rm
as
a
Ser
vice
(P
aa
S
),
a
nd
Softwa
r
e
as
a
Ser
vice
(S
aa
S
).
IaaS
pro
vide
s
sto
rag
e
,
C
PU
s,
net
wor
ks
an
d
oth
e
r
lo
w
-
le
vel
res
our
ces,
Paa
S
pro
vid
e
s
pro
gr
am
m
ing
Gr
a
phic
al
U
se
r
Interface
(GU
I)
,
and
SaaS
pr
ov
i
des pr
e
viou
sly
created a
pp
li
cat
ion
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1179
–
118
6
1180
1.1
.
Pr
ob
le
m
T
he
e
xisti
ng
m
et
hods
work
e
d
to
opti
m
iz
e
the
resou
rce
al
lo
cat
ion
in
cl
oud
as
a
ta
sk
sc
he
du
li
ng
with
QoS c
onstrai
nt
s. Ho
we
ver, th
e com
pu
ta
ti
on
ti
m
e o
f
eac
h
s
ubta
sk i
s call
ed
as a sc
he
du
le
r,
wh
ic
h
is
im
pr
act
ic
al
in
virt
ualiz
ed
cl
oud
syst
em
s.
Ther
e
are
la
r
ge
am
ou
nt
CS
P
are
a
vaila
bl
e
to
offe
r
vi
rtuali
zed
a
nd
dy
nam
ic
resou
rce
on
pa
y
an
d
us
e
basis
in
cl
oud.
How
ever,
t
her
e
are
alm
os
t
CSP
fail
ed
to
m
ai
nta
in
a
qual
it
y
of
se
rv
ic
e
(QoS
)
with
m
ini
m
al
resour
c
e
opti
m
iz
at
ion
.
T
he
sel
ect
io
n
of
se
rv
ic
es
dev
el
op
e
d
to
perform
the
qual
it
y
factors
bec
om
e
s
m
or
e
crit
ic
al
and
c
halle
ngin
g
to
the
ac
hiev
e
m
ent
of
SBSs
,
par
ti
cularly
w
hen
t
he
qual
it
y
factors
are
stric
t.
H
oweve
r
,
e
xisti
ng
m
et
hods
fail
s
to
m
ai
ntain
qu
al
it
y
-
awa
re
serv
ic
e
c
om
po
sit
ion
t
o
enla
r
ge
the
su
ccess
rate
of
com
pu
ti
ng.
T
he
c
urre
nt
m
eth
od
ology
c
on
c
entrated
on
Ia
aS
wh
e
re
cl
oud
se
r
vice
provi
ders
pro
vid
e
d
dissi
m
il
ar
kin
ds
of
res
ources
i
n
the
str
uctu
re
of
VM
occ
urr
ences.
A
n
Iaa
S
pro
vid
e
r
prov
i
des
four
kinds
of
VM
occ
urren
c
es:
sm
all
(S
)
,
m
edium
(M),
l
arg
e
(
L),
a
nd
e
xtra
la
rg
e
(X
L
)
.
S
of
t
war
e
as
a
Se
rv
ic
e
(S
aaS
)
pro
vid
e
rs
giv
es
a
set
of
ap
plica
ti
on
s
util
iz
ing
t
he
Cl
oud
se
rv
ic
es
pro
vid
e
d
by
a
n
I
nfrast
ru
ct
ure
as
a
Ser
vic
e
(
IaaS)
pro
vid
e
r.
The
te
chn
i
qu
e
im
a
gin
es
that
the
IaaS
pro
vid
e
r
pro
vid
es
a
pay
m
ent
on
ly
what
yo
u
util
iz
e
strat
eg
y
on
dem
and
an
d
s
pot
vir
tual
m
achines.
A
Q
oS
-
co
ns
t
raine
d
resou
rc
e
al
locat
ion
m
et
ho
d
introd
uced
to
s
ub
m
it
the
us
er
com
pu
ta
ti
on
ta
sk
in
cl
oud
en
v
iro
nm
ent.
How
ever,
the
m
et
ho
d
is
only
ap
pl
ic
able
for
si
ng
le
VM
instances ty
pe
.
1.2
.
B
ackgro
und
Qu
et
al
[
1]
int
rod
uced
a
n
un
c
ertai
n
-
asse
ssm
ent
-
a
war
e
ince
ntive
te
c
hn
i
que
to
co
ntin
ually
giv
e
ho
nest
assessm
ents
and
prefe
r
giv
in
g
un
ce
rtai
n
ass
e
ssm
ents
ov
er
un
tr
uth
f
ul
or
a
rbi
tra
ry
assessm
e
nts.
Gr
ec
ha
nik
et
al
[2
]
de
sig
ned
a
m
et
ho
d
f
or
P
rovisio
ning
Re
so
urces
with
Ex
per
im
ental
So
ft
war
e
m
odel
li
ng
(P
RE
ST
O)
t
o
i
m
pr
ove
cl
oud
el
ast
ic
it
y
by
le
arn
i
ng
a
nd
re
fining
m
od
el
s
of
un
der
-
c
on
st
raine
d
a
ppli
cat
ion
s
al
l
th
rou
gh
t
he
perform
ance
te
sti
ng
.
Qiu
et
al
[3
]
discu
ssed
a
hiera
rch
ic
al
correla
ti
on
m
od
el
f
or
in
vestig
at
ing
a
nd
eval
uatin
g
correla
te
d
m
ea
su
rem
ents,
w
hich i
nclu
de
d M
arko
v m
od
el
s,
qu
e
uing
the
or
y
, a
nd
a
B
ay
esi
an a
ppr
oac
h.
M
uelde
r
et
al
[4
]
intr
oduce
d
a
vis
ua
ll
y
based
a
nal
ysi
s
ap
proac
h
to
deal
with
c
om
pr
ehensi
on
an
d
analy
zi
ng
t
he
perform
ance &
b
e
hav
i
or of cl
oud
c
om
pu
ti
ng f
ram
ework
s
.
Pal
m
et
al
[5
]
expresse
d
an
ALPINE
,
a
Ba
ye
sia
n
fr
am
ewo
r
k
f
or
cl
oud
perform
ance
a
nd
pr
e
dicti
on.
ALPINE
dep
e
nd
e
d
on
Ba
ye
s
ia
n
Net
wor
ks
(
BNs
)
an
d
c
ont
ai
ned
Cl
oud.
P
apado
poulos
et
al
[
6]
de
velo
pe
d
a
PEA
S
(
Per
for
m
ance
Eval
uation
fr
am
ework
for
A
uto
-
Scal
ing)
str
uctu
re
for
the
eval
uat
ion
of
a
u
to
-
sc
al
ing
m
et
ho
ds.
Si
ng
h
et
al
[
7]
desc
ribe
d
a
n
opti
m
iz
ed
loa
d
bala
nc
ing
f
ram
ewo
r
k
for
the
cl
oud
by
util
iz
ing
A
ct
ive
-
Cl
us
te
rin
g
(
A
C)
al
gorithm
and
A
nt
Colo
ny
O
pti
m
iz
at
i
on
(
ACO
)
to
m
ini
m
iz
e
the
com
plexity
and
tim
e
reducti
on
f
or
a
cl
ie
nt
re
qu
est
for
datace
nter
.
Ma
hd
i
et
al
[8
]
discu
ssed
the
util
iz
at
ion
of
a
dap
ti
ve
re
place
m
ent
cache
(
ARC)
and
pro
bab
il
ist
ic
con
te
nt
place
m
ent
(P
ROB)
al
go
rithm
s,
w
hich
to
gethe
r
are
know
n
as
zon
e
-
base
d
-
a
dap
ti
ve
rep
la
cem
ent
cache
an
d
pro
ba
bili
sti
c
con
te
nt
placem
en
t
(ZB
-
ARCPR
OB
)
.
Elm
ub
arak
et
al
[9]
enh
a
nce
d per
form
ance b
ased
ra
nk
i
ng m
od
el
to help cli
e
nts for ch
oosin
g
th
e b
est
se
r
vi
ces.
Me
sb
ahi
et
al
[10]
intr
oduce
d
a
perf
or
m
ance
evaluati
on,
and
an
analy
ti
cal
correla
ti
on
betwee
n
al
l
basic
loa
d
balancin
g
al
gorit
hm
s
&
recreated
in
cl
oud
com
pu
ti
ng.
Gad
am
et
al
[11]
ex
pr
es
sed
a
c
om
bin
ed
acce
ss
pro
bab
il
it
y
an
d
data
rate
a
s
a
com
m
on
m
e
tric
for
cel
l
c
onnecti
on.
Ha
n
et
al
[12]
devel
op
e
d
a
traf
fic
load
balancin
g
syst
e
m
strive
t
o
ba
la
nce b
et
wee
n
netw
ork
util
it
i
es,
e.
g.,
t
he
a
ve
rag
e
tra
ff
ic
d
e
li
ver
y
la
te
ncy,
an
d
th
e
gr
ee
n
e
nergy
usa
ge.
I
n
[
13
]
de
picte
d
t
he
pur
po
s
e
of
t
he
cl
oud
pa
ra
dig
m
wh
ic
h
e
nhanc
es
the
use
of
net
wor
k
that
pro
vid
e
d
t
he
ca
pab
il
it
ie
s
of
util
iz
ing
one
node
f
ro
m
anot
her
node
.
It
de
scribe
d
the
loa
d
balancin
g
bet
ween
the
cl
ie
nts
an
d
the
ser
ve
rs.
I
n
[
14
]
highli
ghte
d
t
he
tra
deoff
for
offloa
din
g.
T
he
work
pro
vid
e
d
arc
hi
te
ct
ur
e.
The
ge
netic
al
gorithm
integrated
m
ob
il
e
cl
oud
com
pu
ti
ng
f
or
the
pur
pose
of
autom
at
ic
of
f
loading
in
e
nha
ncin
g
the
syst
e
m
resp
onse
ti
m
e.
In
[15
]
ex
plaine
d
loa
d
bala
ncing
play
s
vital
ro
le
in
cl
oud
perform
ance
and
it
s
sta
bili
ty
.
It
dis
cusse
d
var
i
ou
s
load
bala
ncin
g
al
gorithm
s
wh
i
ch
helpe
d
i
n
distribu
ti
ng
t
he
l
oad
am
on
g
the
nodes
and f
ounde
d w
hich suit
ed
the
m
os
t.
1.3
.
O
bj
ec
tiv
es
The pa
per
obje
ct
ives
are
f
ollo
wing as
:
a)
To
de
velo
p
a
r
eso
ur
ce
al
locat
ion
f
ram
ewo
r
k
that
can
av
oid
ov
e
rloa
d
in
th
e
fr
am
ewo
r
k
e
ff
ic
ie
ntly
wh
il
e
m
ini
m
iz
ing
the
num
ber
of se
r
ver util
iz
at
ion
.
b)
To
im
ple
m
ent
the
e
ff
ect
ive
Re
qu
e
st
-
Re
spo
nse
m
od
el
f
or
im
pro
ving
the
c
om
pu
ta
ti
on
proc
ess
&
re
du
ce
th
e
traff
ic
of clo
ud
d
at
a
cente
r
c)
To
desig
n
a
l
oa
d
pre
dicti
on
al
gorithm
that
can
ca
ptu
re
the
fut
ur
e
resou
rce
usa
ges
of
ap
plica
ti
on
s
acc
ur
at
e
ly
without l
oo
king in
side t
he V
Ms.
d)
To
re
duce
CPU
util
iz
at
ion
,
Ba
ndwidt
h
Util
iz
at
ion
&
Re
so
urce
Ut
il
iz
at
ion
com
par
e
tha
n
exist
ing
appr
oach
es
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
An
Eff
ic
ie
nt Fr
am
ew
or
k t
o Improv
e
Qo
S of
CSP
us
in
g
E
nh
an
ce
d Mi
nimal
…
(
Ravi Mah
adeva
n
)
1181
2.
METHO
D
The
propose
d
desig
n
e
xp
la
in
s
the
ef
fecti
ve
a
resou
rce
al
lo
cat
ion
f
ram
ewo
r
k
f
or
av
oid
i
ng
ove
rloa
d
in
the
f
ram
ewo
r
k
w
hile
m
ini
m
i
zi
ng
t
he
nu
m
ber
of
se
r
ver
util
iz
at
ion
.
T
he
m
et
ho
d
m
ai
n
obj
ect
iv
e
is
to
optim
iz
e
al
locat
ed
resou
rce,
m
ini
m
iz
e
the
c
om
pu
ta
ti
on
ti
m
e
&
m
ini
m
iz
e
resour
ce
util
iz
at
ion
s.
T
he
pro
posed
m
et
hod
descr
i
bes
the
s
yst
e
m
arch
it
ectu
re
with
im
plem
entat
ion
ste
ps
an
d
pr
opos
e
d
al
gorithm
details.
Fig
ur
e
1
e
xpresses
the
w
orkf
l
ow
of
i
m
ple
m
entat
i
on
process
flo
w i
n detai
ls.
T
he m
et
ho
d i
m
p
roves t
he
QoS
of
CSP
i
n t
he f
eat
ur
es
of
a
m
ulti
-
di
m
ensio
nal
res
ou
rce
.
T
o
overc
om
e
these
probl
e
m
s,
Enh
a
nce
d
Mi
nim
al
Reso
urce
Op
ti
m
i
zat
ion
Ba
sed
Sc
hedul
ing
al
go
rithm
i
s
desig
ne
d
f
or
reso
l
ving
the
VMPA
C
iss
ue
consi
ders
the
pr
ese
nce
of
res
ources
of
m
ulti
ple
cat
egories.
It
prov
i
des
a
uto
m
a
te
d
res
ource
m
anag
em
ent
fr
am
ewo
rk
t
hat
accom
plishes
a
good
sta
bili
ty
&
reli
abili
ty
of
cl
oud
se
r
vices.
I
n
ov
e
rloa
d
av
oida
nce,
the
capa
bili
ty
of
a
PM
sh
oul
d
be
s
uffi
ci
ent
t
o
sat
isfy
the
resour
ce requires of
al
l
VMs
run
ni
ng
o
n
it
.
O
r
el
se,
the
PM
is
overl
oad
e
d
a
nd
can
le
ad
to d
e
grade
d
perform
ance
of
it
s
VMs.
The
pro
posed
te
ch
nique
quantiz
e
the
quantit
y
of
PMs
util
iz
at
ion
w
hic
h
s
houl
d
be
decr
ease
d
as
l
ong
as
t
hey
can
s
ti
ll
satisfy
the
ne
eds
of
al
l
VM
s.
I
dle
PMs
can
be
tu
rn
e
d
off
to
sav
e
com
pu
ta
ti
on
al
util
iz
at
ion
.
T
he
outc
om
es
il
lustrate
d
t
hat
the
pro
posed
m
et
ho
do
l
og
ie
s
e
valu
at
e
the
nea
r
-
op
t
i
m
al
so
luti
on
s
wh
il
e
eff
ect
ively
ca
pt
ur
in
g
t
he
dyna
m
ic
m
ark
et
de
m
and
,
pro
visio
ning
t
he
c
om
pu
ti
ng
re
sourc
es
to
m
at
ch
the
dem
and,
and
ge
ner
at
in
g
hi
gh
re
venue.
T
he
m
et
ho
do
l
og
y
ca
n
ca
pture
t
he
incre
asi
ng
de
velo
pm
e
nt
of
res
ource
us
a
ge
patte
rn
s
an
d
he
lp
dec
rease
t
he
placem
ent
churn
sig
nificantl
y.
I
n
a
dd
it
io
n,
t
he
c
om
pu
ta
ti
on
processin
g
ti
m
e
of
the pr
opos
e
d
te
chn
i
qu
e
is
ver
y
less.
2.1
.
Im
plem
e
nt
ati
on Pre
-
pr
ocessin
g
S
tep
s
2.1
.
1
Cloud Se
rvice Pr
ov
ider
Cl
oud
ser
vice
pro
vid
er
w
ork
s
a
m
ediat
or
be
tween
cl
oud
us
er
an
d
st
or
a
ge
se
rv
e
r
to
de
sign
e
ff
ect
iv
e
request
-
res
ponse
m
od
el
in
cl
oud
e
nvir
on
m
ent.
T
he
CS
P
st
or
es
di
ff
e
ren
t
t
ypes
of
data
in
a
distri
bu
te
d
m
ann
er
on
diff
e
ren
t
s
erv
e
r
s,
w
hich
geog
raphical
ly
cu
rr
e
nt
in
dif
fer
e
nt
places.
Cl
oud
se
rv
ic
e
pro
vid
e
rs
dea
l
with
enter
pr
ise
i
nfra
structu
re,
an
d
i
t
offer
s
scal
abl
e,
protect
io
n
a
nd
co
ns
ist
ent
s
erv
ic
e
for
cl
ou
d
us
ers
with
m
ini
m
al
cost
.
2.1
.
2
Cloud
U
ser
Cl
oud use
r
s
ho
uld re
gister
as
d
at
a
owne
r wit
h
t
heir
b
a
sic
a
nd c
re
den
ti
al
i
nfor
m
at
ion
detai
ls t
o
get the
log
in
acce
ss.
Hen
ce
,
cl
ou
d
us
er
can
co
ntri
bu
te
t
he
file
or
inf
or
m
at
ion
on
dep
l
oym
ent
serv
e
r
for
a
pp
l
ic
at
ion
us
ers
. T
he u
ploa
ded file
s
will
b
e st
or
e
d
in
a c
loud st
or
a
ge
se
rv
e
r
.
Figure
1.
Wo
r
kfl
ow
of
Pro
po
s
ed
E
nhance
d M
ini
m
al
Resou
rce
Opti
m
iz
ation
Base
d Sch
edu
li
ng
Algo
rithm
in
Cl
oud
Clo
u
d
Service
Prov
id
er
Au
th
en
ticatio
n
p
rocess
Res
o
u
rce
id
en
tif
icatio
n
Select
reso
u
rce
Res
o
u
rce
su
b
scrip
tio
n
Verify
with
CSP
View Sta
tu
s
(active,
p
rocess
in
g
)
Un
d
er
p
rocess
in
g
View
allo
cated
VM
Au
th
en
ticatio
n
pro
cess
View list of
clo
u
d
us
ers
View list of
reso
u
rces
View list of
ap
p
licatio
n
s
Dep
lo
y
app
licatio
n
P
ay
m
en
t
&
creden
tial
done
View list of
run
n
in
g
statu
s
Ch
eck
p
ay
m
en
t
an
d
creden
tial
Do
m
ain
crea
tio
n
Ap
p
ly
E
MROS
Clo
u
d
Server
Ap
p
licatio
n
d
ep
lo
y
m
en
t
VM
allo
cat
io
n
No
Yes
No
Yes
No
Yes
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1179
–
118
6
1182
2.1
.
3
VM Re
s
ou
rce
Alloc
at
i
on
s
The
m
et
ho
dol
og
y
per
i
od
ic
al
ly
execu
te
s,
t
o
est
im
at
e
the
res
ource
al
lo
cat
ion
sta
tus
base
d
on
t
he
cal
culat
ed
upc
om
ing
res
ourc
e
re
quirem
ent
of
VMs.
It
sp
e
ci
fies
the
se
rv
e
r
is
ove
rloa
ded
an
d
the
refor
e
a
f
e
w
VMs
t
urnin
g
on
it
s
houl
d
be
m
igrated
a
w
ay
.
T
he
st
rategy
in
dicat
es
t
he
res
ource
ut
il
iz
at
ion
bey
on
d
the
thres
ho
l
d.
It
de
scribes
a
s
er
ve
r
as
if
the
uti
li
zat
ion
of
t
he
whole
of
it
s
r
eso
ur
ces
is
under
a
t
hr
es
hold.
T
he
fr
am
ewo
r
k
sp
e
ci
fies
the
se
rv
e
r i
s
m
os
tl
y
idle
an
d
pote
ntial
c
and
i
date
t
o s
wi
tc
h
off t
o
save
com
pu
ta
ti
on
powe
r
.
Th
ough,
t
he
m
et
hodo
l
og
y
do
es
co
ns
e
qu
e
ntly
on
ly
wh
e
n
t
he
a
ver
a
ge
res
ource
util
iz
at
i
on
of
whole
a
ct
ively
util
iz
ed
ser
ve
rs
in
t
he
fr
am
ew
ork
a
re
un
der
a
com
pu
ti
ng
t
hresh
old.
A
ser
ver
is
act
ively
util
iz
ed,
a
nd
it
has
at
le
ast
o
ne
V
M
r
unning,
or it i
s
inact
ive.
2.1
.
4
VM Mi
tigatio
n
The
m
et
ho
dolog
y
is
ar
rang
ed
by
a
li
st
of
inte
rn
et
re
so
urce
in
th
e
fr
am
ewo
r
k
in
desce
ndin
g
tem
per
at
ur
e
.
A
n
obj
ect
ive
of
t
he
m
et
ho
dolo
gy
is
possible
to
rem
ov
e
al
l
int
ern
et
res
ource
or
possi
ble
kee
p
thei
r
tem
per
at
ur
e
as
low.
For
e
ve
r
y
server,
fi
rst,
choose
the
V
M
sh
ou
l
d
be
m
igrated
a
way
.
I
f
t
he
VM
is
m
igrated
away,
li
st
of
V
Ms
arr
a
ng
e
d
by
based
on
t
he
ou
t
-
com
ing
te
m
per
at
ur
e
of
th
e
serv
e
r.
T
he
st
rategy
m
ai
nly
fo
cu
se
d
on
m
igrati
ng
a
way
th
e
m
os
t
of
t
he
VM.
F
or
e
ver
y
VM
pr
ese
nted
in
the
li
st,
to
acc
omm
od
at
e
the
desig
ns
o
bs
er
ve
if m
et
ho
do
l
og
y
can disc
ov
e
r
a
d
est
inati
on s
er
ver.
2.1
.
5
Sc
alabi
li
ty a
n
d Fle
xibil
ity
Scal
abili
ty
and
fle
xib
il
it
y
repr
esented
by
the
perform
ance
of
data
ce
ntre.
N
ow,
it
exte
nd
s
t
he
globall
y
routin
g
thr
ou
gh
cl
ie
nt
base
d
on
one
area
da
ta
centres
to
cl
ie
nt
age
nt
of
a
no
t
her
a
rea
data
centres
.
Pro
pose
d
m
et
ho
dolo
gy
c
al
culat
es
inter
ne
t
traf
fic
is
rou
ti
ng
a
ppr
oxim
a
te
ly
the
w
or
l
d
by
init
ia
ti
ng
ap
pro
pr
ia
te
broa
dcast
la
te
ncy
an
d
da
ta
broa
dcast
de
la
ys
for
distribu
te
d
c
om
pu
ti
ng
ap
plica
ti
on.
It
is
ad
diti
onal
ly
serv
in
g
al
lo
t
m
ent
requests
for
l
oa
d balanci
ng rule
.
2.2
.
Enh
an
c
e
d M
ini
m
al
Re
so
urce
Op
timi
z
at
ion B
as
ed
Scheduli
n
g Algo
ri
th
m
En
han
ce
d
Mi
nim
al
Re
so
ur
ce
O
pti
m
iz
at
ion
Ba
sed
Sc
he
du
l
ing
Algo
rithm
is
desi
gn
e
d
t
o
r
ep
rese
nt
fr
am
ewo
r
ks
c
om
po
sed
of t
hous
a
nd
s
of reso
ur
ces
, a
nd it
cr
eat
es possible t
o
re
present
both
ph
ysi
cal
and
virtu
a
l
resou
rces
de
ve
lop
in
g
par
ti
cul
ar
cl
oud
ideas
s
uch
as
t
he
i
nfrastr
uctu
re
e
la
sti
ci
t
y.
The
m
od
el
ha
nd
le
s
tra
ff
i
c
m
on
it
or
ing
bet
ween
use
r
bas
es
an
d
data
ce
ntres.
T
he
default
traf
fic
m
on
it
or
i
ng
poli
cy
is
m
app
ing
tr
aff
ic
a
t
nearby
data
ce
ntre
re
ga
rd
i
ng
netw
ork
la
te
nc
y
fr
om
us
er
ba
se.
The
m
od
el
represe
nts
flo
w
le
vel
o
f
a
pp
l
ic
at
ion
with
res
pecti
ve
proce
ssin
g
w
orkf
l
ow
a
nd
in
vo
l
ved
fa
ct
or.
Althou
gh
te
ch
nique
is
ap
plica
ble
la
r
ge
num
ber
of
us
er
bases
.
Th
e
m
et
ho
d
is
c
apab
le
for
ser
ver
s
a
nd
data
centres;
these
kinds
of
syst
e
m
s
req
ui
re
spe
ci
fic
te
chn
iq
ues
.
Th
e
pro
po
se
d
te
c
hn
i
q
ue
s
dis
play
j
ob
sch
ed
uling
le
vel
by
le
ve
l.
Howe
ver,
the
pr
opos
e
d
te
chn
i
qu
e
fo
c
us
es
on
ap
plica
ti
on
t
o
re
trie
ve
e
ff
ic
ie
nc
y
of
the
assi
gn
e
d
jo
b
to
th
e
syst
em
fr
om
va
rio
us
re
gions.
T
he
pro
po
se
d
te
ch
nique
is
handl
ed
by
t
he
proc
ess.
The
c
halle
ng
e
of
the
te
c
hn
i
qu
e
is
to
de
crease
t
he
qua
ntit
y
of
act
ive
ser
ver
s
thr
ough
the
lo
w
loa
d
wit
h
no
sacri
fici
ng
pe
rfor
m
ance
ei
ther
c
urre
ntly
or
in
the
f
utu
r
e.
In
t
he
fr
am
ewo
r
k
are
need
e
d
to
a
void
os
ci
ll
at
ion
.
The
a
ver
a
ge
consum
ption
of
al
l
resour
ce
s
on
act
ive
ser
ver
s
is
invok
e
d
by
pro
po
s
ed
te
ch
niqu
e.
T
he
pr
opos
e
d
te
c
hniq
ue
e
xpress
es
m
any
processes
for
r
edu
ci
ng
c
om
pu
ta
ti
on
tim
e
and
e
nergy
c
on
s
um
ption
i
n
c
om
pu
ti
ng
syst
em
s.
Its
m
easur
e
var
i
ous
Q
oS
at
tri
bute
s
an
d
e
valua
te
th
e
relat
ive r
a
nkin
g of Cl
oud ser
vices.
Her
e
,
CSP
cre
at
es
res
o
urce
i
nfor
m
at
ion
a
nd
al
locat
es
m
e
m
or
y
util
iz
at
io
n.
Cl
oud
us
e
rs
identify
t
he
resou
rces
base
d
on
co
st,
m
e
m
or
y
al
locat
ion
a
nd
pr
ocessi
ng
tim
e
befor
e
finali
zed
the
a
ny
cl
oud
ser
vi
ces.
T
he
resou
rce
is
sel
ect
ed
by
cl
ou
d
us
e
r
f
or
subsc
riptio
n
t
o
dep
l
oy
their
a
pp
l
ic
at
ion
i
n
cl
oud
env
i
ronm
ents.
Cl
oud
us
er
can
de
ploy
the
ap
plica
ti
on
afte
r
their
cre
den
ti
al
&
paym
ent
ve
rificat
ion
by
CSP.
O
nce,
de
plo
ym
ent
proces
s
is
com
plete
d
then
us
e
r
ca
n
vie
w
sta
tus
of
the
serv
e
r,
wh
et
he
r,
ser
ve
r
is
act
ive
or
un
der
processin
g
or
ina
ct
ive.
Af
te
r
dep
l
oym
ent
of
cl
oud
use
r’
s
ap
plica
ti
on,
cl
oud
us
e
r
can
vie
w
t
he,
com
pu
ti
ng
processin
g
tim
e,
CPU
util
iz
at
ion
,
ba
ndwi
dth
util
iz
at
i
on
&
m
e
m
or
y
util
iz
at
ion
.
T
he
ps
e
udo
c
ode
of
pro
posed
al
gorithm
is
exp
l
ai
ned
belo
w
in
d
et
ai
l
s
:
The Inp
ut:
Re
so
urce a
ll
oc
at
ion, clo
ud se
rv
i
ce p
rovide
r (CSP), cl
ou
d user (CU
)
Out
p
ut:
Visu
a
li
ze
the
opti
m
i
zed
loa
d
&
tra
f
fic,
c
om
pu
ti
ng
proces
sin
g
ti
m
e
(CPT),
CP
U
util
iz
at
ion
(
CPUU)
,
band
width uti
li
zat
ion
(BU
)
&
m
e
m
or
y uti
li
za
ti
on
(MU
)
Proced
ure:
Star
t;
Br
ow
se
Porcess
Cl
oud ser
vice
pro
vid
e
r
(C
SP) an
d
Cl
ou
d user a
uth
e
ntica
ti
on
proces
s;
View
the a
vail
able res
ource
with
resp
ect
i
ve
CSP
detai
ls;
I
den
ti
f
y t
he
ap
plica
ti
on r
e
quirem
ent;
Sele
ct
the CS
P
with
serv
ic
e
util
it
y detai
ls;
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
An
Eff
ic
ie
nt Fr
am
ew
or
k t
o Improv
e
Qo
S of
CSP
us
in
g
E
nh
an
ce
d Mi
nimal
…
(
Ravi Mah
adeva
n
)
1183
Alloc
at
e the r
es
ourc
es w
it
h m
e
m
or
y;
Creat
e
u
se
rb
a
se;
Assign
the task;
A
pp
ly
EMR
O
S for A
ppli
cat
i
on d
e
plo
ym
ent;
If
A
pp
li
cat
io
n dep
l
oyed
View St
at
us
of ser
ver (acti
ve
,
pro
ces
sin
g)
Else
View u
nd
e
r
processin
g &
re
-
dep
l
oy the a
pp
li
cat
ion
En
d
I
f
If
VM all
ocate
d
View All
ocate
d VM
Else
Ver
ify
with C
S
P & real
locat
e
VM;
View
li
st
of
a
ppli
cat
ion
s;
View VM
s
run
ning stat
us
;
Disp
la
y
op
ti
m
iz
ed
loa
d
&
traf
fic,
c
om
pu
ti
ng
proc
essin
g
tim
e
(CPT),
CPU
util
iz
at
ion
(CP
UU)
,
band
width uti
li
zat
ion
(BU
)
&
m
e
m
or
y uti
li
za
ti
on
(MU
)
.
3.
RESU
LT
S
A
ND
DI
SCUS
S
ION
3
.1
.
Pr
ogram
mi
ng
S
etu
p
The
pro
posed
m
et
ho
d
is
im
pl
e
m
ented
in
I
nt
el
i6
C
or
e
pr
oc
essor,
with
16
GB
RAM
,
500
GB
Me
m
or
y
with
W
i
ndows
7
Ulti
m
at
e
op
erati
ng
syst
e
m
s.
The
pro
pos
ed
is
im
ple
m
e
nted
i
n
NetB
e
ans
8.0,
J
D
K
(Jav
a
Dev
el
op
m
ent
Kit)
1.8,
M
YSQL
data
base
5.
5,
wit
h
Jel
ast
ic
Cl
oud
se
rver,
in
Ja
va
progra
m
m
ing
en
vir
onm
ent.
The
pro
pose
d
fr
am
ewo
r
k
util
iz
es
Cl
oudS
im
&
iTe
xt
li
br
a
r
y
to
de
plo
y
an
d
vis
ualiz
e
the
opti
m
iz
ed
re
s
ource
resu
lt
.
3
.2
.
In
pu
t
P
ar
amet
er
s
The
in
put
par
a
m
et
ers
are
ex
pl
ai
ned
detai
ls
in
Table
1
to
de
plo
y
the
propose
d
al
gorithm
to
evaluat
e
the ef
fici
ency
of prop
os
e
d
m
et
hodo
l
og
y.
Table
1
.
Cl
oud Ex
per
im
ental
Ev
al
uatio
n pa
r
a
m
et
er D
et
ai
ls
Para
m
eters
Valu
e
Userbas
e
06
Reg
io
n
06
Datacenter
4
(DC1
and
DC2
)
Virtual
Machin
e
2
5
(DC
1
)
5
0
(DC
1
)
7
5
(DC
3
)
&
7
5
(
DC
4
)
Data Cent
re
V
M
Xen
Nu
m
b
e
r
o
f
Pr
o
cess
Ma
ch
in
e W
ise
16
Data Cent
er
Pr
o
ce
ss
in
g
Speed
1
0
0
M
IPS
Data Cent
re
V
M
P
o
licy
Ti
m
e
Shared
Data Cent
re
OS
W
in
d
o
ws 7
VM
Me
m
o
ry
2048
Data Cent
re
Ar
ch
itectu
re
X8
6
Ban
d
wid
th
1
0
0
0
M
b
p
s
3
.
3
.
E
xp
eri
m
ent
al R
es
ult
The
propose
d
m
et
ho
d
e
xpla
ins
the
e
valuati
on
m
at
rix
to
c
al
culat
e
the
ef
fici
ency
an
d
ut
il
iz
ation
of
pro
po
se
d
m
echan
ism
co
m
par
e
tha
n
e
xisti
ng
a
ppr
oach.
T
he
pro
posed
al
gorit
hm
is
evaluate
d
on d
iffe
re
nt
t
ypes
of
in
put
pa
ram
et
ers
to
fin
d
out
reli
abili
ty
an
d
ef
fici
ency.
T
he
pro
pose
d
te
chn
i
qu
e
is
e
val
uated
with
f
ollow
i
ng
par
am
et
ers
na
m
el
y
Co
m
pu
ta
ti
on
P
ro
c
essin
g
Tim
e,
Ba
ndwidth
Util
iz
at
i
on
&,
CPU
Ut
il
iz
at
ion
and
R
eso
ur
ce
Util
iz
at
ion
in d
et
ai
ls.
3.
3
.
1
Co
m
pu
t
at
i
on
Pr
ocessi
ng
Ti
me
(
CPT
)
The
Com
pu
ta
ti
on
P
rocessi
ng
Ti
m
e
(CPT)
c
om
pu
te
s
the
ti
m
e
co
nsum
ption
to
process
the
us
er
r
eq
uest
from
data
center
in
cl
ou
d
en
vi
ronm
ent.
Hence
,
it
processes
for
retrie
ve
the
requested
quer
y
fr
o
m
databas
e.
The
com
pu
ta
ti
on
pr
ocessin
g
Tim
e
is
the
rati
o
bet
ween
t
he
data
request
an
d
ba
ndwidt
h
co
nsu
m
pt
ion
pe
r
use
r.
T
he
CPT is cal
culat
ed wit
h
m
at
hem
at
ic
al
ex
pr
es
sion i
n
E
quat
io
n (1)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1179
–
118
6
1184
P
e
r
u
s
e
r
ss
c
o
p
u
t
p
r
o
c
e
BW
D
T
(1)
Wh
e
re
Tcom
pu
tp
ro
ces
s
is
t
ot
al
com
pu
ta
ti
on
ti
m
e,
D
is
r
equ
e
ste
d
data
&
B
WPeruse
r
is
total
ba
ndwidt
h
util
iz
at
ion
u
se
r
w
ise
3.
3
.
2
B
an
dwi
d
th
Ut
il
iz
at
ion
The
Ba
ndwidt
h
util
iz
at
ion
is
total
net
work
util
iz
at
ion
to
de
plo
y
the
a
pp
li
cat
io
n.
T
he
ba
ndwidt
h
util
iz
at
ion
is
expresse
d
in
E
q
uation
(
2)
to
c
al
culat
e
the
re
qu
e
st
-
res
ponse
m
od
el
.
Ba
nd
width
c
ons
umpti
on
is
a
Total
all
ocated
b
a
ndwidt
h div
ided by tota
l
num
ber
of
us
er
re
qu
e
sts.
r
t
o
t
a
l
p
e
r
u
s
e
r
N
BW
BW
(2)
3.
3
.
3
CPU U
ti
liz
at
ion
CPU
util
iz
at
ion
e
xpresse
s
t
he
co
nsum
ption
of
physi
cal
re
s
ources
f
or
s
pec
ific
ta
sk
.
CP
U
Util
iz
at
ion
is
div
isi
on
of
A
ve
rag
e
per
i
od
of
ba
ck
gro
und
ta
sk
(
Id
le
ta
sk)
without
loa
d
by
ave
rag
e
per
i
od
of
backg
rou
nd
ta
s
k
with loa
d. T
he C
PU
util
iz
at
ion
is cal
c
ulate
d i
n
E
qu
at
io
n (
3) to e
valuate e
ffi
ci
ency of pro
p
os
e
d
m
et
ho
ds
.
PU
Ut
il
izat
io
n
=
Avg
Period
of
B
a
c
kgr
ou
n
d
Ta
sk
w
ith
out
L
oad
Avg
Perio
d
of
B
a
c
kgr
oun
d
Ta
sk
w
ith
L
oa
d
×
100
(3)
3.
3
.4
Mem
ory R
es
ou
rce
Util
iz
at
ion
The
m
e
m
or
y
resour
ce
util
iz
at
ion
re
pr
ese
nts
the
ho
w
pro
pose
d
syst
em
is
ef
fecti
ve
t
o
m
ini
m
iz
e
the
m
e
m
or
y
util
iz
at
ion
.
T
he
m
e
m
or
y res
ource
ut
il
iz
at
ion
is
cal
c
ulate
d
a
s
t
he
ba
sed
on
t
he
subtract
io
n
of
al
locat
ed
m
e
m
or
y
fr
om
buff
e
rs
a
nd
ca
ched
m
e
m
or
y.
The
m
e
m
or
y
resour
ce
util
iz
at
ion
is
cal
culat
ed
m
at
he
m
atical
ly
in
E
quat
ion (
4).
Utili
za
t
ion
=
Allocate
d
Mem
m
or
y
−
B
uff
e
rs
Mem
oer
y
−
C
a
c
he
d
Mem
or
y
Alloca
t
e
d
Mem
or
y
×
100
(4)
Table
2
dis
play
s
the
Com
pu
ta
ti
on
Processi
ng
Ti
m
e
(CPT),
B
andwidt
h
Util
iz
at
ion
,
CPU
Util
iz
at
ion
(CP
U
U)
&
Mem
or
y
Re
source
Util
iz
at
ion
s
(MRU
)
for
giv
e
n
i
nput
pa
ram
et
ers.
The
pro
po
se
d
te
ch
n
iq
ue
is
eval
ua
te
d
on
giv
e
n
e
valuati
on
pa
ram
et
ers
with
Acti
ve
M
on
it
ori
ng
(A
M
)
[
16]
,
Fair
Ro
und
R
obin
(F
R
R)
[
17]
,
Ro
und
Ro
bin
(RR)
[
18]
exi
sti
ng
a
ppr
oa
ch
es.
Acc
ordi
ng
to
Ta
ble
2,
it
no
ti
ced
t
hat
an
E
nhan
ced
Mi
nim
a
l
Re
so
urc
e
Op
ti
m
iz
ation
Ba
sed
Sc
hedul
ing
(EMR
OS)
has
the
best
sc
or
e
on
eve
ry
r
especti
ve
c
onstrai
nt
f
or
giv
e
n
inputs
par
am
et
ers
.
Table
2
.
Disp
la
ys
the Com
pu
t
at
ion
P
r
ocessin
g
Tim
e (CPT)
,
Ba
ndwidt
h Uti
li
zat
ion
, CP
U Uti
li
zat
ion
(
CP
UU)
& Mem
or
y R
esource
Util
iz
at
i
on
s
(
MR
U)
Lear
n
in
g
Algo
rithm
s
CPT(
m
s)
BU
(M
b
p
s)
CPUU (%
)
MRU (
%)
FRR
2
0
4
9
.4
4
55
19
22
RR
6
2
0
2
.7
7
50
18
28
AM
6
0
5
3
.5
1
50
17
17
EM
ROS
1
7
4
7
.7
2
30
12
14
Accor
ding
to
F
igure
2
t
o
4
pe
rfor
m
ances,
it
ob
s
er
ved
that
Pr
op
os
e
d
EM
ROS
s
how
s
good
re
su
lt
bes
t
on
CP
T,
BU
,
a
nd
CP
U
U
&
M
RU
eval
uation
m
at
rix
on
giv
e
n
in
pu
t
par
am
et
ers.
I
n
te
rm
s
of
CPT
(C
om
puta
ti
on
processi
ng
tim
e),
FRR
(F
ai
r
-
Roun
d
Ro
bin)
is
cl
os
est
te
ch
ni
qu
es
to
Propos
ed
EMR
OS
.
H
ow
e
ve
r,
t
he
FR
R
fail
s
to
optim
iz
ed
t
he
res
ource
a
nd
predict
the
l
oad.
Be
half
of
CPU
util
iz
at
io
n,
Ba
nd
widt
h
Util
iz
at
ion
&
Mem
ory
Re
so
urce
Util
iz
at
ion
(MRU
),
AM
(A
ct
ive
Mon
it
ori
ng)
is
the
cl
os
est
m
e
thod.
But,
it
unable
to
offer
eff
ic
ie
nt
Re
qu
est
-
Re
spo
ns
e
Mo
del
an
d
scheduli
ng
po
li
cy
to
m
ini
m
i
ze
the
com
pu
ta
ti
on
proc
essin
g
tim
e
and
op
ti
m
iz
ed
the
physi
ca
l
re
so
urces
.
P
rop
os
e
d
al
gorithm
m
a
intai
ns
eq
ui
valent
w
orkl
oa
ds
on
al
l
the
a
vaila
ble
VMs
and
the
qu
a
ntit
y
of
requests
prese
ntly
assigne
d
to
V
M.
EMR
OS
a
vo
i
ds
ov
e
rloa
d
in
the
f
ram
ewo
r
k
ef
fecti
vely
wh
il
e
m
ini
m
iz
in
g
the
quantit
y
of
se
rv
e
rs
us
e
d.
Pro
po
s
ed
te
ch
niqu
e
i
m
pr
oves
the
CPT
301.7
2
m
illi
secon
ds,
BU
20
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
An
Eff
ic
ie
nt Fr
am
ew
or
k t
o Improv
e
Qo
S of
CSP
us
in
g
E
nh
an
ce
d Mi
nimal
…
(
Ravi Mah
adeva
n
)
1185
Mbp
s
,
CP
UU
5%
&
MR
U
3%
on
gi
ven
i
nput
par
am
et
ers
co
m
par
e
than
e
xisti
ng
m
et
ho
dolog
y.
Finall
y,
it
cl
aim
s
that p
rop
os
ed
EMR
OS
m
et
ho
dolo
gy is
best
on all res
pecti
ve
c
onstrai
nts
Figure
2.
Com
pu
ta
ti
on
P
r
oce
ssing Ti
m
e (CPT)
i
n
m
il
l
ise
c
onds
(m
s)
Figure
3.
Ba
nd
width Uti
li
zat
ion (M
bp
s
)
Figure
4.
CPU
Util
iz
at
ion
(
B
U)
an
d
Me
m
ory
Resource
Util
iz
at
ion
(
MR
U
)
in
Per
ce
ntag
e
4.
CONCL
US
I
O
N
The
pa
per
pr
e
sents
En
ha
nce
d
Mi
nim
al
Re
so
urce
O
ptim
i
zat
ion
base
d
Sche
du
li
ng
Al
gorithm
to
reducin
g
the
r
eso
ur
ces
a
nd
m
ai
ntaining
th
e
QoS.
T
he
pro
posed
m
et
ho
ds
av
oid
s
ov
erloa
d
f
or
res
ource
al
locat
ion
,
&
util
iz
ed
optim
i
ze
res
ources.
The
fr
am
ewo
r
k
c
om
pu
te
s
th
e
res
ource
util
i
zat
ion
base
d
on
cl
ie
nt
requirem
ent.
The
e
ntire
util
iz
at
ion
pu
rpos
e
is
to
e
nh
a
nc
e
the
QoS
of
CSP
in
the
a
tt
ribu
te
s
of
a
m
ul
ti
-
dim
ension
al
re
so
urce.
T
he
propose
d
m
et
ho
dolo
gy
util
iz
ed
op
ti
m
iz
ed
ro
ut
ing
to
decr
e
ase
the
traff
ic
in
a
cl
oud
env
i
ronm
ent.
The
f
ram
ewo
r
k
al
so
helps
cl
oud
us
er
to
prefer
best
CS
P
accor
ding
to
th
ei
r
pr
i
or
se
rv
ic
es.
The
m
od
el
rep
rese
nt
s
flow
le
vel
of
app
li
cat
io
n
wi
th
res
pecti
ve
processin
g
wor
kfl
ow
a
nd
in
vo
l
ve
d
fact
or.
Alth
ough
te
chn
iq
ue
is
ap
plica
ble
la
r
ge
nu
m
ber
of
use
r
bas
es
.
T
he
m
et
hod
is
ca
pa
ble
for
se
r
ver
s
an
d
data
cente
rs;
these
kinds
of
syst
e
m
s
req
uire
sp
e
ci
fic
te
ch
nique
s.
Its
m
easur
e
va
rio
us
Q
oS
at
tribu
te
s
a
nd
evaluate
t
he
re
la
ti
ve
rankin
g
of
Cl
oud
se
r
vices.
The
pro
po
se
d
te
chn
iq
ues
dis
play
job
sche
duli
ng
le
vel
by
le
v
el
.
H
owev
er,
th
e
pro
po
se
d
te
ch
ni
qu
e
f
oc
us
es
on
a
ppli
cat
ion
to
retrie
ve
e
ff
ic
i
ency
of
the
assi
gn
e
d
jo
b
to
the
syst
e
m
fr
om
va
rio
us
reg
i
on
s
. Pr
opose
d
te
c
hn
i
qu
e
im
pr
ov
es t
he
C
PT
301.7
2
m
ill
ise
conds, BU
20 M
bp
s
, CP
U
U 5%
& MR
U
3% o
n
giv
e
n
in
put p
ar
a
m
et
ers
com
par
e tha
n
e
xisti
ng m
et
ho
dolo
gy
In
f
uture,
the
pa
per
ca
n
be
ex
te
nd
e
d
t
o
ap
pl
y
the
pri
vacy
of
cl
ie
nt
ap
plica
ti
on
,
use
r
lo
g
a
nd
act
ivit
y
without
C
SP
di
scl
os
ure.
F
or
im
pr
ov
in
g
data
analy
ti
cal
pro
cess,
H
DF
S
ca
n
be
integ
rated
to
exec
ute
t
he
ta
s
k
eff
ect
ive
w
ay
i
n
cl
ou
d
e
nv
i
ronm
ent
.
2
0
4
9
.
4
4
6
2
0
2
.
7
7
6
0
5
3
.
5
1
1
3
4
7
.
7
2
0
2
0
0
0
4
0
0
0
6
0
0
0
8
0
0
0
FR
R
RR
AM
EMR
OS
CPT
55
50
50
30
FR
R
RR
AM
EMR
OS
0
20
40
60
BU
0
10
20
30
19
18
17
12
CP
U U
tili
z
ati
on
MRU
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
1179
–
118
6
1186
REFERE
NCE
S
[1]
Qu,
L.,
W
ang,
Y.,
&
Orgun
,
M.,
“
A
novel
in
ce
nt
ive
m
ec
h
anis
m
for
trut
hful
per
form
anc
e
assess
m
ent
s
of
cl
o
ud
servic
es”
,
In
Pr
oce
ed
ings
of
th
e
2016
Inte
rnat
ional
Confe
r
ence
on
Aut
onomo
us
Age
nts
&
M
ult
iage
n
t
Syst
ems.
Inte
rnational
Fo
undati
on
f
or A
ut
onomous
Age
nts
and
Mul
ti
agen
t Syste
ms
,
pp
.
132
5
-
1326,
2016
.
[2]
Grec
hani
k
,
M.
,
Luo,
Q.
,
Pos
h
y
v
an
y
k
,
D
.
,
&
Port
er,
A
.
,
“
Enha
ncing
rule
s
for
c
lou
d
resourc
e
provi
sioning
via
l
ea
rn
ed
software
p
erf
or
m
anc
e
m
odel
s”
,
In
Proc
ee
dings
of
th
e
7th
ACM/SPE
C
on
In
te
rn
ati
onal
Confe
re
nce
on
Pe
rform
ance
Engi
ne
ering
,
pp.
209
-
214,
2016.
[3]
Qiu,
X.,
Dai,
Y.,
Xiang,
Y.
,
&
Xi
ng,
L
.
,
“
A
hie
r
ar
chi
c
al
cor
relati
o
n
m
odel
for
ev
aluati
ng
r
el
i
abi
l
ity,
per
form
ance,
a
nd
power
consum
pt
ion
of
a
c
loud
se
rvic
e
”,
IE
EE
Tr
ansacti
ons
on
Sy
stems,
Man,
and
Cyb
erne
tics:
Sy
stems
,
vol
.
46
,
n
o.
3,
pp
.
401
-
412
,
2016.
[4]
Muelde
r,
C
.
,
Zh
u,
B
.
,
Ch
en,
W
.
,
Zh
ang,
H.
,
&
Ma,
K.
L
.
,
“
Visual
an
aly
sis
of
c
loud
computing
per
form
anc
e
usi
ng
beha
vior
al
li
n
es”
,
IE
EE t
rans
act
i
ons on
vi
sual
izat
ion
and
computer
gra
phic
s
,
vol
.
22,
no
.
6
,
pp
.
16
94
-
1704,
2016
.
[5]
Palm,
E
.
,
Mi
tra,
K.,
Saguna
,
S
.
,
&
Åhlund
,
C.
,
“
A
Ba
y
esi
an
S
y
st
em
for
Cloud
Per
form
anc
e
Diagn
osis
and
Predi
ct
i
on”,
In
Cloud
Compu
ti
ng
Te
chnol
ogy
and
Scienc
e
(
CloudCom)
,
2016
IEE
E
Inte
rnat
ion
al
Conf
ere
nc
e
,
p
p.
371
-
374
,
201
6.
[6]
Papadopoul
os,
A.
V.
,
Al
i
-
El
d
in
,
A.
,
Årzé
n
,
K.
E.
,
Tordsson,
J.
,
&
E
lmroth,
E.,
“
PEAS
:
A
per
fo
rm
anc
e
evalua
t
i
o
n
fra
m
ework
for
aut
o
-
sca
li
ng
stra
te
gi
es
in
cl
oud
appl
i
ca
t
ions”,
A
CM
Tr
ansacti
ons
on
Mode
li
ng
and
Pe
rform
anc
e
Ev
aluation o
f
C
omputing
Sy
st
ems (
T
OMPECS)
,
vol.
1
,
no
.
4
,
pp
.
1
-
39,
2016
.
[7]
Singh,
R
.
,
&
Praka
sh,
S.,
“
Enh
anc
ement
of
Re
source
Al
loc
a
ti
o
n
using
Loa
d
B
al
an
ci
ng
in
Clo
ud
Com
puti
ng”
,
Inte
rnational
Jo
urnal
of Adv
an
c
ed
R
ese
arch
in Com
pute
r Sc
i
en
ce
,
vol
.
8
,
no
.
5
,
pp.
11
-
17
,
2017
.
[8]
Mahdi, A
.
S., &
Muni
y
and
ih, R
.
C.
,
“
Enha
nc
eme
nt
of
Cloud
Perf
orm
anc
e and
Sto
rag
e
Consum
pti
on
Us
ing
Adap
tive
Repl
a
ce
m
ent
C
ac
he
and
Prob
abi
li
st
ic
Cont
en
t
Pla
ce
m
ent
Al
gorit
hm
s”,
Journal
of
Theoret
i
cal
and
Appl
i
e
d
Information
Tec
hnology
,
vol
.
84
,
no.
3,
pp.
376
-
3
84,
2
016
.
[9]
El
m
ubar
ak,
S.
A
.
,
Yous
if,
A.,
&
Bashir,
M.
B
.
,
“
Perform
anc
e
-
bas
ed
Ranki
ng
Model
for
C
loud
Sa
aS
Servic
es”
,
I
.
J
.
Information
Tec
hnology
and
Co
mputer
Scienc
e
,
vol.
1
,
pp
.
65
-
71
,
2017
[10]
Mesbahi,
M.
R.
,
Hashem
i,
M.
,
&
Rahmani,
A.
M.
,
“
Perform
anc
e
e
val
ua
ti
on
and
an
aly
s
is
of
loa
d
ba
l
anc
ing
al
gori
thms
in
cl
oud
comput
ing
envi
ronm
ents
”,
In
W
eb
Re
se
arch
(
ICWR)
,
2
016
Se
cond
In
ter
nati
onal
Con
fe
r
enc
e
,
pp.
145
-
15
1,
2016
[11]
Gada
m
,
M.
A
.
,
Ng,
C.
K.,
Nordi
n,
N.
K.,
Sali,
A
.
,
&
Hashim
,
F.
,
“
Hy
brid
Chann
e
l
Gain
Acc
ess
Cel
l
As
socia
ti
on
f
or
Loa
d
B
al
an
ci
ng
in
Dow
nli
nk
L
TE
-
Advanc
ed
H
et
Nets”
,
In
Computer
and
Com
municat
ion
Enginee
ring
(
ICCCE
)
,
2016
Inte
rnat
ion
al
Conf
ere
nce
,
p
p.
337
-
342
,
201
6.
[12]
Han,
T.,
&
Ans
a
ri,
N.,
“
A
tr
aff
ic
loa
d
b
al
an
ci
ng
f
ramework
for
so
ftwa
re
-
def
ine
d
r
adi
o
acce
ss
n
et
w
orks
powere
d
b
y
h
y
brid
en
erg
y
so
urc
es”
,
IE
EE
/
AC
M
Tr
ansacti
ons
on
Net
work
ing
(
TO
N
)
,
vol. 24, n
o.
2
,
pp
.
1038
-
1
051,
2016
[13]
Ta
l
asil
a
,
S.
,
Va
ni,
H.
,
Sa
i,
K
.
,
Mani,
D.
,
Krish
na
Redd
y
,
V.
,
"
Loa
d
B
al
an
ci
ng
Techni
ques
f
or
Eff
i
ci
en
t
Tr
aff
i
c
Mana
gement
in
Cloud
Env
ironment",
Int
ernati
on
al
Journal
o
f
Elec
tri
cal
and
Co
mputer
Eng
ine
er
ing
(
IJE
CE)
,
Vol
.
6,
No.
3,
pp.
963
-
973,
2016
.
[14]
Nadim
Akhtar
,
N.
Sriniv
asa
n,
"
Notic
e
of
Ret
r
action
In
te
rm
it
t
entl
y
Connecte
d
Clo
udle
t
S
y
st
em
to
Ob
ta
in
an
Opt
imal
Offloa
ding
Pol
icy
",
Indone
sian
J
ournal
of El
e
ct
ri
cal
Engi
ne
ering and Inf
orm
ati
cs
(
IJE
EI)
,
Vol.
4
,
No.3,
2016
.
[15]
Ravi
T
eja
Kan
a
kal
a
,
V.
,
Krish
na
Redd
y
,
V.
,
"P
erf
orm
anc
e
A
naly
s
is
of
Loa
d
Balanci
ng
Tec
hnique
s
in
Clou
d
Com
puti
ng
Environm
ent
",
TEL
KOMNIKA
Indon
esian
Journal
of
Elec
tri
cal
Enginee
ring,
Vol.
13
,
No.
3,
pp
.
568
-
573,
2015
.
[16]
Aw
ada
ll
a
,
M.
H.
A.
,
“
Heuri
sti
c
Approa
ch
for
Schedu
li
ng
De
pende
nt
Re
al
-
Tim
e
Ta
sks
”
,
Bullet
in
o
f
Elec
tri
c
al
Engi
ne
ering
and
Informatic
s
,
vol
.
4
,
no
.
3
,
pp
.
21
7
-
230,
2
015
.
[17]
Jam
m
al
amada
ka
,
S.
K.
R
.
,
Duvv
uri,
K.
B
.
K.
,
Ch
,
D.
A.
,
&
Padm
ini
,
P.
,
“
Buil
din
g
Fault
To
ll
r
ence
withi
n
C
louds
at
Network
Le
v
el”,
Inte
rnat
ional
Jo
urnal
of El
e
ct
ri
c
al
and
Comput
er
Engi
n
ee
ring
,
vo
l.
6
,
no
.
4
,
pp
.
15
60
-
1569,
2016
.
[18]
Li
,
J.,
Meng,
X.,
W
en,
J
.
,
&
Xu,
Y.,
“
An
improve
d
m
et
hod
of
SV
M
-
BP
S
O
fea
tur
e
select
ion
base
d
on
cl
oud
m
odel
”
,
Indone
sian J
our
nal
of
Elec
tric
al
Engi
ne
ering
and
Computer
Sc
ie
n
ce
,
vol
.
12
,
no
.
5
,
pp
.
3979
-
3986
,
2014
Evaluation Warning : The document was created with Spire.PDF for Python.