Int
ern
at
i
onal
Journ
al of Inf
orm
at
ic
s
and
Co
m
munic
at
i
on
Tec
hn
olog
y (IJ
-
I
CT)
Vo
l.
6
,
No.
3
,
D
ece
m
ber
201
7
, pp.
179
~
188
IS
S
N:
22
52
-
8776
,
DOI: 10
.11
591/ijict
.
v6
i
3
.
pp
179
-
188
179
Journ
al h
om
e
page
:
http:
//
ia
esj
ou
r
nal.co
m/
on
li
ne/in
dex
.php
/
IJ
ICT
An Agen
t Archit
ectur
e for
QoS
-
ba
sed Web
Servi
ce
Compositi
on Usin
g t
he
Skylin
e
Algorith
m
El
-
Alami
A
youb
*
, Hair
Abdel
latif
La
bora
tor
y
of
Applie
d
m
at
hemat
ic
s a
nd
Sci
ent
if
i
c
Ca
lc
ul
at
ion
,
Fa
cul
t
y
of
Sci
ence
s a
nd
T
ec
hno
log
y
,
Sult
an
Moul
a
y
Slim
ane
Univer
s
ity
,
B.
P.
523,
Be
ni
Mel
la
l
,
Moroc
co
.
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
A
ug
12
th
,
20
1
7
Re
vised
Oct
20
th
, 201
7
Accepte
d
Nov
26
th
,
20
1
7
W
eb
servic
e
co
m
positi
on
is
a
conc
ep
t
base
d
on
the
bui
lt
of
an
abstr
act
proc
ess,
b
y
co
m
bini
ng
m
ult
ipl
e
exi
sting
class
insta
nce
s,
wher
e
during
the
exe
cu
ti
on,
each
servic
e
class
is
re
place
d
b
y
a
con
cre
t
e
servi
ce,
sel
ec
t
ed
from
seve
ra
l
web
se
rvic
e
c
andi
da
tes
.
Thi
s
appr
oa
ch
has
as
an
adva
nta
g
e
gene
ra
ti
ng
fl
exi
b
le
and
low
coupling
appl
i
ca
t
ions,
base
d
on
it
s
conc
eption
on
m
an
y
elem
ent
ar
y
m
odule
s
avai
la
bl
e
on
the
w
eb.
Th
e
proc
ess
of
servic
e
sele
c
ti
on
during
the
compos
it
ion
is
base
d
on
seve
ra
l
ax
es,
one
of
the
se
ax
es
is
the
QoS
-
base
d
web
servic
e
sele
c
ti
on.
Th
e
Qos
or
Quali
t
y
of
Servic
e
re
pre
sent
a
set
o
f
par
amet
ers
th
at
cha
r
acte
ri
ze
the
non
-
func
t
iona
l
web
servi
c
e
aspe
ct
(e
x
ec
u
ti
o
n
ti
m
e,
cost,
e
tc...).
Th
e
compos
it
ion
of
web
servic
es
base
d
on
Qos
,
is
the
proc
ess
which
al
l
ows
the
sele
ct
io
n
of
the
web
se
rvic
es
tha
t
fulfi
ll
the
user
n
ee
d,
b
ase
d
on
i
ts
qual
ities.
S
e
lect
ed
services
should
opti
m
ize
the
glob
al
QoS
of
the
compos
ed
proc
ess,
whi
le
sati
sf
y
ing
al
l
the
constraints
spec
ified
b
y
the
cl
i
ent
in
al
l
QoS
par
amete
rs
.
In
thi
s
pap
er,
we
propose
a
n
appr
oac
h
base
d
on
th
e
conc
ep
t
of
ag
ent
s
y
st
e
m
and
Sk
y
li
ne
appr
oac
h
to
eff
ectivel
y
sel
ect
services
for
c
om
positi
on,
and
re
duci
ng
th
e
num
ber
of
ca
ndid
at
e
serv
ices
to
be
gene
r
ated
and
conside
r
ed
in
treat
m
ent.
To
eva
lu
ate
our
appr
oac
h
ex
pe
rimentall
y
,
we
use
a
sev
era
l
ra
ndom
dat
ase
ts
of
servic
e
s
with
ra
ndom
v
alues of
qu
al
i
ti
es.
Ke
yw
or
d:
Mult
i
-
Ag
e
nt S
yst
e
m
Qos
Sk
yl
ine
Web Se
rv
ic
e
Web Se
rv
ic
e C
om
po
sit
ion
Copyright
©
201
7
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights
reserv
ed
.
Corres
pond
in
g
Aut
h
or
:
El
-
Alam
i Ayou
b,
Lab
or
at
ory
of
Applie
d
m
at
hem
at
ic
s an
d
Sci
entifi
c Cal
culat
ion
,
Faculty
of S
ci
e
nces a
nd Tec
hnol
og
y,
Su
lt
an
M
oula
y Sl
i
m
ane Univ
ersit
y,
B.P. 523
, Beni
Me
ll
al
, Mo
ro
c
co.
Em
a
il
: ay
ou
bala
m
i6@g
m
ai
l.c
om
1.
INTROD
U
CTION
Du
e
to
t
he
pe
r
petual
increa
se
of
we
b
ser
vic
es
with
sim
il
a
r
functi
onal
it
y,
the
com
po
sit
ion
process
beco
m
es
costly
in
te
r
m
s
of
respon
se
ti
m
e,
t
ake
into
co
ns
i
der
at
io
n
the
num
ber
of
can
di
date
serv
ic
es
in
eac
h
cl
ass
of
the
c
om
po
sit
ion
.
F
or
this
pur
pose
s
ever
al
wor
ks
a
nd
resea
rches
hav
e
pro
pose
d
m
et
ho
ds
t
o
s
ol
ve
this
pro
blem
of
co
m
po
sit
ion
ba
se
d
on
the
Q
os.
These
m
et
ho
ds
al
low
the
sel
ect
ion
of
a
pa
rtic
ular
se
rv
ic
e
from
a
m
on
g
se
ve
ral
cand
i
date
ser
vices,
f
or
eac
h
do
m
ai
n
or
cl
ass
of
th
e
co
m
po
sit
ion
,
in
the
ai
m
to
generate
a
com
po
sit
ion
w
it
h
a b
et
te
r
qua
li
ty
o
f
com
po
si
ti
on
.
Using
the
e
xh
austive
sea
rch,
can
be
us
e
fu
l
to
fin
d
the
be
st
com
bin
at
ions
of
se
r
vices,
wh
ic
h
they
hav
e
t
he
opti
m
al
le
vels
of
Q
oC.
But
this
m
et
hod
of
searc
h
sti
ll
no
n
-
pra
ct
ic
al
,
with
a
v
ery
la
r
ge
nu
m
ber
of
po
s
sible
c
om
bin
at
ion
s
that
c
an
be
ge
ner
at
e
d
a
nd
te
ste
d
in
e
ver
y
e
xec
ution
.
T
his
c
omplexit
y
of
e
xhaustiv
e
search
r
el
at
ing
to
the
nu
m
ber
of
cl
asses
in
a
com
po
sit
ion
,
al
so
to
the
num
ber
of
serv
ic
e
cand
idate
s
in
each
cl
ass.
So,
this
pro
blem
can
be
m
od
el
ed
as
a
com
bin
at
or
ia
l
pr
oble
m
with
NP
-
ha
rd
c
omplexit
y.
Am
ong
the
us
ef
ul m
et
ho
ds t
o
s
olv
e t
his c
om
po
sit
ion
pro
blem
, is to
us
e
the S
kyli
ne
sel
ect
ion
tec
hn
i
que. T
his tech
nique
of
Sk
yl
ine
ai
m
s
t
o
re
duce
the
num
ber
of
can
di
date
ser
vices
in
each
com
po
s
it
ion
cl
ass,
a
nd
therefo
re
re
du
ces
the
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
8776
IJECE
V
ol.
6
,
No.
3
,
Decem
ber
20
1
7
:
179
–
188
180
total
nu
m
ber
of
com
bin
at
ions
.
This
reducti
on
is
accom
plished
by
the
el
i
m
inati
on
of
al
l
do
m
inate
d
ser
vices,
and
keep
i
ng
dom
inant
on
es,
wh
e
re
we
say
that
on
e
ser
vic
e
is
do
m
inate
d
by
ano
the
r,
if
at
le
as
t
on
e
of
its
qu
al
it
ie
s is sm
al
l t
han
a
nothe
r
se
rv
ic
e.
In
t
his
pap
e
r,
we
hav
e
f
ocus
ed
our
wor
k
on
us
in
g
a
n
a
ge
nt
-
based
m
et
h
od,
in
orde
r
t
o
searc
h
a
nd
perform
com
p
os
it
ion
s
be
twe
en
tw
o
S
kyli
nes
ser
vices,
w
her
e
each
S
kyli
ne
c
on
ta
ins
th
e
dom
inant
cand
i
dat
e
serv
ic
es
of
a
cl
as
s.
The
ro
le
of
age
nts
is
to
ge
ner
at
e
a
ne
w
Sk
yl
ine
by
the
gen
e
rati
on
of
a
par
ti
al
com
po
sit
ion
from
two
pr
e
vi
ou
s
cl
asses
,
a
nd
kee
ping
j
ust
the
do
m
inant
c
om
po
sit
ion
s,
a
nd
rem
ov
in
g
the
rest
of
the
s
erv
ic
e
cand
i
dates.
Th
e
m
et
ho
d
c
on
si
st
to
rep
eat
t
he
sam
e
op
er
at
io
n
betwee
n
t
he
par
ti
al
create
d
com
po
sit
ion
,
a
nd
th
e
fo
ll
owin
g
cl
as
s
unti
l
the
e
nd
of
the
com
posit
ion
,
t
o
ob
ta
i
n
finall
y
a
S
ky
li
ne
inclu
des
the
insta
nces
of
the
op
ti
m
al
co
m
bin
at
ions
of
t
he m
ai
n
com
po
sit
ion
.
2.
QOS
-
BASE
D CO
MPO
SITI
ON PR
OCES
S
A
ser
vice
cl
ass
is
def
ine
d
by
a
set
of
we
b
se
rv
ic
es
that
pr
ovide
the
sam
e
functi
onal
it
y,
bu
t
they
have
pro
bab
ly
dif
fere
nt
non
-
f
unct
ion
al
c
har
act
e
risti
cs
(d
if
fer
e
nt
values
of
Q
oS).
In
our
w
ork
we
will
co
nsi
der
a
cl
oud
c
om
pu
ti
ng
in
wh
ic
h,
di
ff
ere
nt
we
b
se
rv
ic
es
a
re
dep
l
oye
d,
an
d
that
con
ta
in
a
cl
ass
ific
at
ion
of
se
r
vices
accor
ding
to
their
f
unct
iona
l
char
act
erist
ic
s
(class
of
serv
ic
es
),
al
s
o
a
desc
riptio
n
of
non
-
f
unct
ion
al
char
act
e
risti
cs
for
each
se
rv
ic
e
(Qos).
This
descr
i
ption
of
functi
onal
and
non
-
functi
ona
l
serv
ic
es
are
store
d
and
publishe
d
in
a
serv
ic
e
re
gistry
(UDDI)
,
wh
ic
h
is
acce
ssible
via
the
web,
an
d
m
eet
custom
er
need
s
f
or
si
m
ple
web
ser
vices
a
nd
com
po
sit
io
n
se
r
vices.
O
ur
obj
ect
ive
is
to
create
a
process
in
t
he
cl
oud
us
i
ng
s
yst
e
m
agen
t t
o be a
bl
e to sati
sfy c
ust
om
e
r’
s co
m
posit
ion
r
e
quest
s.
2.1. Qo
S
p
ar
am
eters
The q
ualit
y of
serv
ic
e
pa
ram
e
te
rs
is a set
of
non
-
f
un
ct
io
nal
qu
a
ntit
at
ive cha
racteri
sti
cs,
w
hich
determ
ines and
d
esc
ribes
the
perform
ance o
f
a w
e
b ser
vice.
Th
ese
p
a
ram
eter
s ca
n
incl
ude
v
a
rio
us
att
rib
ut
es,
li
ke
pri
ce,
repu
ta
ti
on
, a
vaila
bi
li
ty
, r
el
ia
bili
ty
,
r
es
pons
e
tim
e, b
a
ndwidt
h,
t
hroug
hput,
et
c.
All t
hese
pr
e
vi
ou
s
par
am
et
ers
can
b
e e
valuate
d wit
h real
posit
ive
or n
e
gative
values
. For p
osi
ti
ve
par
am
et
ers,
servic
e c
us
to
m
ers
sh
oul
d
sea
rch to m
axi
m
iz
e the
m
, s
uc
h
as
re
pu
ta
ti
on a
nd a
vaila
bili
ty
, an
d f
or
ne
gative
pa
ram
et
ers,
it
s nee
d
to
be
m
ini
m
iz
ed
su
ch
as c
os
t a
nd r
es
pons
e
tim
e. For bett
er
pr
es
entat
ion
we wi
ll
w
ork
just
with
neg
at
ive
par
am
et
ers,
for t
he posi
ti
ve
pa
ram
et
ers,
they will
b
e tra
nsfo
r
m
ed
into
ne
gat
iv
e b
y m
ulti
plyi
ng
t
heir val
ue
s
by
-
1.
We s
uppos
e the
vecto
r Q
s
= {q
1
(s), ...,
q
n
(s)
}
to
r
e
pr
ese
nt Qos val
ues of we
b
se
r
vice
S,
a
nd q
i
(s)
determ
ines the
values
of the
i
-
th att
rib
ute of t
he
se
rv
ic
e
S.
2.2. Qo
S
c
alcu
lation f
or
c
om
po
site
ser
vices
The
QoS
par
a
m
et
ers
for
co
m
po
sit
e
serv
ic
es
or
com
po
si
ti
on
are
cal
cu
la
te
d
base
d
on
the
Q
oS
par
am
et
ers
of
it
s
com
po
nen
t
serv
ic
es
.
Co
ns
i
der
i
ng
C
=
{
S1
,
S2,
...,
S
n}
as
a
com
po
sit
ion
of
the
f
ollow
i
ng
web
ser
vices:
{S1,
S
2,
...,
Sn}
,
the
qual
it
y
of
c
om
po
sit
ion
C
is
de
f
ine
d
a
s
Q
(C)
=
{
q
1
(C
),
q
2
(C
),
…
,
q
n
(C)}
,
wh
e
re
q
i
(C
)
is
est
i
m
at
ed
valu
e
of
i
-
t
h
at
tri
bute
,
a
nd
wh
ic
h
can
be
cal
cul
at
ed
us
in
g
a
n
aggre
gation
f
unct
ion.
This
functi
on
aim
s
to
agg
regat
e
the
values
of
co
rresp
ondi
ng
Qo
s
at
trib
ut
es
of
al
l
com
p
on
e
nt
ser
vices.
Us
ual
aggre
gation
functi
ons
f
or
Q
oS
cal
culat
ion
a
re
su
m
m
ation
,
m
ul
ti
plica
ti
on
,
and
m
ini
m
iz
ati
on
f
un
ct
io
n
[
1]
.
Th
e
Table
1 give a
n
e
xam
ple o
f
s
om
e Q
oS
att
ri
bute
s a
nd it
s aggr
e
gatio
n
f
unct
ion
.
Table
1.
E
xam
ples
of
Q
oS
a
ggre
gatio
n
f
unct
ion
s
Ty
p
e
Attribu
tes
Fu
n
ctio
n
s
Su
m
m
atio
n
Res
p
o
n
se ti
m
e
,
Price
q
(C)
=
∑
Rep
u
tatio
n
q
(C)
=
∑
Multip
licatio
n
Av
ailab
ility
q
(C)
=
∏
Mini
m
u
m
Thro
u
g
h
p
u
t
q
(C)
=
m
in
q(S
i
)
2.2. Pro
blem
s
tatemen
t
QoS
-
base
d
ser
vice
com
po
sit
ion
is
a
pro
ble
m
,
wh
ic
h
ai
m
s
at
find
i
ng
t
he
serv
ic
e
c
om
bin
at
ion
t
hat
m
axi
m
iz
es the Q
oS
values
of
a com
po
sit
ion.
The
sim
ple
m
e
thod
for
fin
ding
the
op
ti
m
al
com
bin
at
ion
is
to
us
e
a
n
ex
ha
us
ti
ve
s
earc
h,
to
ge
ner
at
e
and
com
par
e
a
ll
po
ssi
ble
co
m
bin
at
ion
s
of
cand
i
date
w
eb
serv
ic
es
.
F
or
a
n
e
xam
ple
com
po
sit
ion
re
quest
wi
t
h
N
cl
asses
an
d
M
serv
ic
es
pe
r
cl
ass,
there
ex
ist
N
M
po
ssible
com
bin
at
ion
s
to
com
par
e.
T
her
e
fore,
the
c
os
t
of
perform
ing
a
n
exh
a
us
ti
ve
s
earch
ca
n
be
ver
y
ex
pen
si
ve
in
te
rm
s
of
c
om
pu
ta
ti
on
tim
e
and
m
e
m
or
y
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
ICT
IS
S
N:
22
52
-
8776
An
A
ge
nt Arc
hi
te
ct
ur
e for
Qos
-
B
as
ed
We
b S
ervi
ce Com
pos
it
ion
…
(
El
-
Ala
mi Ay
oub
)
181
occupati
on,
ne
eded
t
o
sto
r
e
al
l
gen
erate
d
com
bin
at
io
n.
Co
ns
e
quent
ly
,
Exh
a
us
ti
ve
search
m
et
ho
d
is
inap
pro
pr
ia
te
f
or
r
un
-
ti
m
e
serv
ic
e
com
po
sit
ion
in
syst
e
m
s
with
dynam
ic
needs
[
1].
In
the
fo
ll
owin
g
s
ect
ion,
we
will
pro
pos
e
a
so
luti
on
f
or
this
pro
blem
,
based
on
the
sel
ect
ion
an
d
c
om
po
sit
ion
of
web
se
r
vices,
us
in
g
do
m
inance
rel
at
ion
s
hip
betw
een
ca
nd
i
date
serv
ic
es
i
n
ea
ch
ser
vice
cl
a
ss,
to
sel
ect
s
kyli
ne
ser
vice
s.
The
pro
po
se
d
s
ol
u
ti
on u
s
e the
m
ult
i agen
t
pa
rad
i
gm
.
3.
SKY
LI
NE
A
ND M
ULTI
-
A
GENT S
YS
T
EM FO
R QO
S
-
BA
SED
COMPO
SITIO
N
The
pur
po
se
of
Q
oS
-
base
d
c
om
po
sit
ion
is
to
sel
ect
a
set
of
serv
ic
es
,
one
from
each
cl
ass
of
ser
vices,
in
suc
h
a
w
ay
that
the
qu
al
it
y
of
t
he
sel
ect
ed
c
om
bin
at
ion
b
e
m
axi
m
iz
e
d.
Nam
ely
that,
the
s
el
ect
ion
of
t
he
op
ti
m
al
cand
id
at
e
serv
ic
e
f
rom
each
cl
ass,
do
e
s
no
t
pro
vi
de
necessa
rily
an
op
ti
m
al
com
po
sit
ion
.
T
he
refor
e
,
to
fin
d
the
c
orrect
so
luti
on,
oth
e
r
com
bin
a
ti
on
s
of
se
rv
ic
es
need
t
o
be
consi
der
e
d.
O
n
the
oth
er
ha
nd,
w
e
sh
oul
d
m
ention
that
not
al
l
s
erv
ic
es
a
re
pote
ntial
can
did
at
es
to
ge
ner
at
e
the
s
olu
ti
on.
C
on
s
eq
ue
ntly
,
the
idea
of
our
m
et
ho
d
is
to
us
e
the
sk
yl
ine
m
e
tho
d
t
o
disti
ngui
sh
betwee
n
t
he
can
did
at
e
s
erv
ic
es
f
or
a
giv
e
n
com
po
sit
ion
,
a
nd
no
t
pote
nti
al
on
es
f
or
ea
ch
cl
ass.
The
sk
yl
ine
is
a
m
et
hod
that
aim
to
reduce
the
search
sp
ace
of
a
give
n
set
of
el
em
e
nts,
by
us
i
ng
t
he
dom
inance
relat
ion
s
betwe
en
ser
vices
bas
ed
on
Q
oS
values,
t
o
identify
an
d
el
im
inate
serv
ic
es
that
are
dom
i
nated
by
oth
e
r
serv
ic
es
in
t
he
sam
e
c
la
ss
[1
]
.
A
ser
vice
A
is
sai
d
do
m
inate
d
by
ano
t
her
ser
vice
B,
if
B
is
bette
r
tha
n
or
eq
ua
l
to
A
i
n
al
l
at
trib
utes,
a
nd
B
is
stric
tl
y
bette
r
in
at
le
ast
o
ne
Qos a
tt
ribu
te
.
Defi
niti
on
[
1].
(Dom
inance)
Con
si
der
a
ser
vice
cl
ass
S
,
a
nd
tw
o
ser
vice
s
X,
Y
∈
S,
c
ha
racteri
zed
by
a
set
of
Q
of
Q
oS
at
tribu
te
s.
X
do
m
inate
s
y,
de
no
te
d
as
X
≺
Y,
if
X
is
as
good
or
bette
r
than
Y
in
al
l
par
am
et
ers
in
Q
an
d
bette
r
in
at
le
ast
on
e
pa
ram
eter
in
Q,
∀
k
∈
[
1,
|Q|]
:
q
k
(X)
≤
q
k
(Y)
and
∃
k
∈
[
1,
|
Q|]
: q
k
(
X) <
q
k
(Y).
Defi
niti
on
[
1].
(S
kyli
ne
Se
r
vices)
The
s
kyli
ne
ser
vices
of
a
serv
ic
e
cl
ass
S,
com
pr
ise
th
os
e
ser
vices
in S
t
hat are
no
t do
m
inate
d by any
oth
e
r
se
rvi
ce, {x
∈
S|¬
∃
y
∈
S: y
≺
x}.
Figure
1. Me
as
ur
i
ng the
Dista
nce
to
the
Sk
yl
ine
Figure
1
s
how
s
an
exam
ple
of
a
pp
li
cat
ion
of
the
S
kyli
ne
m
et
ho
d
f
or
a
giv
e
n
cl
ass,
ta
kin
g
int
o
consi
der
at
io
n
t
wo
Q
os
pa
ram
et
ers
to
e
valuat
e
each
se
r
vice,
cost
a
nd
exec
ution
tim
e.
Each
po
i
nt
of
the
gra
ph
represe
nts
a
c
and
i
date
ser
vi
ce,
with
the
coor
din
at
es
of
a
point
c
orr
esp
onding
to
the
val
ues
of
tho
s
e
par
am
et
ers.
U
s
ing
t
he
m
et
ho
d
of
s
kyli
ne
al
lo
wing
to
re
duce
the
optim
al
serv
ic
es
by
disti
nguis
h
dom
inant
and
do
m
inate
d
ser
vices,
the
dom
inant
ca
nd
i
date
serv
ic
es
i
n
blu
e
c
olor,
a
nd
the
se
r
vice
s
do
m
inate
d
in
or
a
nge
colo
r.
Ta
king
the
exam
ple
of
A,
B,
C
an
d
D
serv
ic
es
,
we
se
e
that
B,
C
and
D
al
l
do
m
inated
by
the
se
rv
i
ce
A
,
because
al
l
Q
os
Attrib
utes
of
serv
ic
e
A,
a
re
bette
rs
(less
)
than
Qos
at
tribu
te
s
of
ot
her
serv
ic
es
.
Ta
kin
g
t
he
case
of
se
r
vices
E
an
d
A,
we
can
obse
rv
e
t
ha
t
serv
ic
e
A
is
bette
r
tha
n
E
on
pri
ce,
but
it
is
infer
i
or
t
han
E
on
execu
ti
on
tim
e
at
tribu
te
,
in
th
is
case
we
sai
d
that
the
se
r
vi
ces
E
a
nd
A
a
re
inc
om
par
abl
e
ser
vices,
a
nd
both
belo
ngs to
the
op
ti
m
al
sk
yl
ine ser
vices.
3.1. Si
mp
le
S
k
yline co
mp
os
it
ion
The
idea
of
the
Sim
ple
Sk
yl
ine
Com
po
sit
ion
SSC
m
eth
od
is
to
ta
ke
adv
a
ntag
e
of
reducin
g
th
e
nu
m
ber
of
ca
nd
i
date
se
rv
ic
es
to
be
co
nsi
der
e
d
in
c
ompo
sit
io
n,
by
usi
ng
the
Sk
yl
i
ne
m
et
ho
d
on
each
com
po
sit
e
cl
as
s.
Af
te
r
el
im
in
at
ing
of
do
m
ina
te
d
serv
ic
es
,
the
SSC
pe
rform
s
an
exh
au
sti
ve
searc
h
to
ge
ner
at
e
al
l
po
ssi
ble
c
om
bin
at
ion
just
from
the
optim
al
sk
yl
ine
se
rvi
ces,
be
fore
pe
rfor
m
ing
a
gain
a
S
kyli
ne
qu
e
ry
on
p
rice
Execu
tio
n
ti
m
e
d
’execu
tio
n
A
B
C
D
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
8776
IJECE
V
ol.
6
,
No.
3
,
Decem
ber
20
1
7
:
179
–
188
182
gen
e
rated
c
ombinati
ons
to
ge
t
the
op
ti
m
al
sk
yl
ine
ser
vic
es
as
il
lustrate
d
in
fig
ur
e
1.
This
m
et
ho
d
a
i
m
s
at
acce
le
rate
the
com
po
sit
ion
,
a
lso
en
surin
g
t
he
a
bili
ty
of
a
pp
li
cat
io
n
to
f
ind
t
he
opti
m
al
com
po
sit
ion,
just
a
m
on
g t
he se
le
ct
ed
ser
vice ca
nd
i
dates.
Figure
2. Exa
m
ple o
f
com
posit
ion
of
3 ser
vi
ce cl
asses
Figure
3.
E
xa
m
ple o
f
ser
vic
es r
e
pr
ese
nted
by tw
o qu
al
it
y at
tribu
te
s
Con
si
der
t
he
e
xam
ple
of
Fi
gure
2,
w
he
re
a
com
po
sit
ion
r
equ
e
st
is
proce
ssed,
with
thre
e
com
po
sit
e
serv
ic
e
cl
asses
A,
B
and
C,
and
desc
ribe
d
by
two
Q
o
S
par
am
et
ers
P
rice
and
E
xec
ution
ti
m
e.
Fi
gure
3
represe
nt
the
r
esult
of
first
ste
p
of
exec
utio
n
of
SSC
al
gor
it
h
m
.
The
SSC
perform
s
a
sk
yl
ine
req
ue
st
on
each
serv
ic
e
cl
ass
of
c
om
po
sit
ion
,
to
reduce
t
he
nu
m
ber
of
ca
ndidate
se
r
vices
to
c
om
bin
ed,
represe
nted
i
n
blu
e
colo
r.
Com
par
i
ng
t
he
nu
m
ber
o
f
ca
ndidate
c
om
bin
at
ion
s
to g
ene
rate
in
eac
h
case, w
e find
that
SSC
al
gor
it
h
m
gen
e
rate
4*
5*5=10
0
can
did
a
te
com
bin
at
ion
s.
O
n
the
oth
e
r
ha
nd,
to
perf
or
m
an
exh
a
ust
ive
search
withou
t
SSC w
e
m
us
t g
ener
at
e
13*1
4*14
=
2584 c
om
bi
nations.
Figure
4. Me
as
ur
i
ng the
Dista
nce to t
he Skyl
ine
The
sec
ond st
ep of
S
SC al
gor
it
h
m
is p
erform
ing
a n
e
w
s
ky
li
ne
qu
e
ry on
the g
e
ner
at
e
d
c
om
bin
at
ions
,
to
sel
ect
fi
nally
the b
est
a
nd o
ptim
al
sk
yl
ine
com
po
sit
ions
.
Eac
h
blu
e
poi
nts
in
fi
gure 4
,
r
ep
rese
nts
a
n
optim
a
l
com
po
sit
ion
,
com
po
sed
f
rom
three
se
rv
i
ces
bel
ongs
A,
B
an
d
C
serv
ic
e
cl
asse
s
.
T
he
Sim
ple
Sk
yl
ine
Com
po
sit
ion
i
s
a
sim
ple
m
eth
od
ai
m
s
to
pe
rfor
m
ing
e
xhaustive
sea
rc
h
on
a
re
duce
s
pa
ce
of
se
rv
ic
es
,
an
d
m
ai
ntain the sa
m
e set of
opti
m
al
co
m
bin
at
ion
s
.
Ser
vice
Cl
ass A
Ser
vice
Cl
ass B
Ser
vice
Cl
ass C
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
ICT
IS
S
N:
22
52
-
8776
An
A
ge
nt Arc
hi
te
ct
ur
e for
Qos
-
B
as
ed
We
b S
ervi
ce Com
pos
it
ion
…
(
El
-
Ala
mi Ay
oub
)
183
Al
go
rit
hm
1
:
S
i
m
ple Sk
yl
ine
Com
po
sit
ion
Inp
ut
:
a
set
of
servi
ce cl
as
ses
Comp.
Ou
tp
ut:
a
set
of
o
ptim
al servi
ces S
o.
1: for all
Cl
s
∈
Comp
do
2:
S
sky
.ad
d( S
kyl
in
e(
Cl
s
)
)
3: en
d
f
or
4: So
= S
sky
.
pop()
5: for all
c
∈
S
s
ky
do
6:
So = co
m
bin
e(
So, c)
7: en
d
f
or
8: So
= Skyl
in
e(
So
)
3.2.
I
mpr
ov
e
d
Skyli
ne
co
mp
os
itio
n
In
order
to
i
m
pr
ov
e
the
pe
rfor
m
ance
of
the
pr
e
vious
al
gorithm
of
com
po
sit
ion
.
We
pro
pose
a
m
et
ho
d
of
co
m
po
sit
ion
bas
ed
on
sk
yl
ines
sel
ect
ion
and
agen
t
syst
e
m
par
a
dig
m
,
in
su
c
h
a
way
that,
the
app
li
cat
io
n
of
sk
yl
ine
will
be
perform
ed
after
each
c
om
bin
at
ion
of
tw
o
s
erv
ic
e
cl
asses
,
and
t
he
re
su
lt
of
t
his
sk
yl
ine
set
,
will
be
com
bin
e
d
again
with
t
he
fo
ll
owin
g
cl
ass
of
com
po
sit
io
n.
T
hese
t
wo
operati
ons
of
bi
nar
y
com
po
sit
ion
a
nd
s
kyli
ne
reducti
on
m
us
t
be
rep
eat
ed
unt
il
the
la
s
t
serv
ic
e
cl
ass
of
the
g
lob
al
com
po
sit
ion
fig
ur
e
5.
This
appr
oach
re
du
ce
m
or
e
and
m
or
e
the
total
num
ber
of
can
di
date
com
bin
at
ion
s
,
and
at
th
e
sa
m
e
tim
e, it keep
th
e sam
e resu
lt
s o
f
the
pr
ece
di
ng alg
ori
thm
s.
Figure
5. Gl
obal
p
rese
ntati
on
of im
pr
ov
e
d
s
ky
li
ne
com
po
sit
ion
Figure
6. Exa
m
ple o
f
ap
plica
ti
on
of im
pr
oved
s
kyli
ne
co
m
po
sit
ion
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
8776
IJECE
V
ol.
6
,
No.
3
,
Decem
ber
20
1
7
:
179
–
188
184
To
im
ple
m
ent
our
ap
proac
h,
we
pro
pose
a
n
al
gorithm
based
on
t
he
noti
on
of
syst
em
agen
ts,
t
o
ben
e
fit
f
ro
m
their
ca
pa
bili
ties
in
te
rm
s
of
sync
hron
iz
at
i
on
of
ta
s
ks
a
nd
par
al
le
li
sm
.
This
im
ple
m
e
ntati
on
requires
the
de
finiti
on
of
tw
o
functi
on
al
it
ie
s,
the
bin
a
ry
co
m
bin
at
ion
and
sk
yl
ine
reduct
ion
,
us
in
g
an
a
gen
t
-
or
ie
nted
a
rc
hitec
ture.
I
n
t
his
con
te
xt,
t
wo
a
gen
ts
has
be
en
pro
posed
to
a
ccom
plish
thes
e
ta
sk
s:
c
om
posit
ion
agen
t
a
nd
c
ontrol
ag
ent.
Fi
gur
e
7
a
nd
8
s
how
an
overall
work
i
ng
dia
gra
m
and
an
act
ivit
y
diagr
am
,
wh
i
c
h
ind
ic
at
es the
tasks a
nd acti
viti
es p
e
rfor
m
ed
by
each
a
gen
t
.
Figure
7. dem
on
strat
io
n of re
duct
ion an
d bi
na
ry com
po
sit
io
n
in
a c
om
po
sit
ion
proces
s
The
id
ea
of
the
ap
proach
is
t
o
ta
ke
a
dv
a
ntag
e
of
the
c
ollec
ti
ve
intel
li
gen
t
of
syst
e
m
agent
s
to
res
olve
this
com
po
sit
ion
pro
blem
.
At
the
first
ste
p,
a
set
of
init
ia
l
com
po
sit
ion
a
gen
ts
gen
e
rate
d,
w
he
re
each
agen
t
is
accor
ding
to
one
co
ncr
et
e
se
rv
ic
e
of
the
first
ser
vice
cl
ass
of
c
om
po
sit
ion
.
A
com
posit
ion
age
nt
can
be
init
ia
li
zed
with
a
vecto
r
of
Q
oS
at
trib
utes
of
a
c
orrespo
nd
i
ng
we
b
se
rv
ic
e
or
with
t
he
ve
ct
or
QoS
of
a
par
ti
al
com
po
sit
ion
.
Af
te
r
t
he
creat
ion
of
al
l
age
nt
s
of
fir
st
cl
ass,
they
sen
d
an
ACL
m
essages
with
their
vec
tor
of
q
ualit
ie
s
to
t
he
co
ntr
ol
age
nt.
Af
te
r
recei
ving
al
l
m
essages,
the
c
on
t
ro
l
a
ge
nt
trig
gers
the
p
r
ocess
of r
ed
uc
ti
on
,
by
ap
plyi
ng
th
e
Sk
yl
ine
al
gorithm
on
the
r
ecei
ved
data
(
QoS
at
tribu
te
s
)
to
disti
nguis
h
the
dom
inant
an
d
do
m
inate
d
age
nts.
I
n
orde
r
to
reduce
t
he
c
and
i
dates
age
nt
s,
the
con
tr
ol
agen
t
sen
d
an
A
gree
m
ess
age
to
do
m
inant
a
gent
s
an
d
a
C
an
c
el
m
essage
to
oth
e
rs.
When
a
com
po
sit
ion
age
nt
recei
ve
s
an
A
gree
m
essage
from
the
con
tr
ol
age
nt,
the
a
gen
t
sta
rts
ope
rati
on
of
re
pro
du
ct
io
n.
This
operati
on
ai
m
s
t
o
c
reate
a
new
set
of
com
po
sit
ion
a
gen
ts
,
c
orres
pondin
g
to
t
he
serv
ic
es
of
t
he
f
ollow
i
ng
se
rv
ic
e
cl
ass
,
w
her
e
each
new
age
nt
init
ia
li
zed
with
the
Q
os
vect
or
of
a
web
se
r
vice
of
fo
ll
owin
g
cl
ass,
an
d
the
Qo
s
vecto
r
of
i
ts
creator
a
ge
nt.
T
o
cal
culat
e
his
own
Q
os
Vect
or,
a
c
om
po
sit
io
n
a
ge
nt
use
a
ggre
gatio
n
funct
ion
s
to
cal
culat
e
ne
w
Q
os
at
tri
bu
te
s
,
from
the
Qo
s
Vecto
r
of
it
s
own
ser
vice,
a
nd
the
Q
os
vect
or
of
it
s
ow
n
c
reator
age
nt.
A
fter
the
creati
on
of
al
l
new
c
om
po
sit
ion
a
ge
nts
of
th
e
fo
ll
owin
g
cl
a
ss,
the
creat
o
r
com
po
sit
ion
a
gen
t
sto
ps
it
s
e
xecu
ti
on.
I
n
th
e
case
wh
e
re
t
he
c
ompo
sit
io
n
a
ge
nt r
ecei
ving
a
C
ance
l
m
essage,
the
age
nt
st
op
s
it
s
exec
ution. O
nce
al
l
com
po
sit
io
n
agen
ts
of
ol
d
ge
ner
at
io
n
ha
ve
stop
pe
d
their
execu
ti
on,
t
he
new
c
om
po
sit
ion
a
gen
ts
sta
rt
s
the
co
m
m
un
i
cat
ion
with
the
co
ntr
ol
agen
t
by
se
nd
i
ng
a
gain
th
ei
r
Qo
s
vecto
r
.
Th
os
e
operat
ion
s
of
re
duct
ion
a
nd
reprod
uction
rep
eat
e
d
unti
l
arr
ivi
ng
at
the
la
st
serv
ic
e
cl
ass
of
c
om
po
sit
ion.
The
dom
in
ant
com
po
sit
io
n
age
nts
rem
ain
ed
a
t
la
st
represents
the
s
olu
ti
on
of
ser
vice
com
posit
ion.
T
he
de
scriptio
n
of
co
m
po
sit
ion
age
nt
an
d
c
ontr
ol
agen
ts
can
be
s
umm
ar
iz
ed
as
fo
ll
ows
:
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
ICT
IS
S
N:
22
52
-
8776
An
A
ge
nt Arc
hi
te
ct
ur
e for
Qos
-
B
as
ed
We
b S
ervi
ce Com
pos
it
ion
…
(
El
-
Ala
mi Ay
oub
)
185
Figure
8. Acti
vi
ty
d
ia
gr
am
o
f c
om
po
sit
ion
a
nd c
on
t
ro
l a
ge
nts
Contr
ol
agen
t
:
is
the
c
om
po
ne
nt
res
ponsi
ble
f
or
c
omm
u
nicat
ion
with
a
ll
com
po
si
ti
on
s
age
nts.
It
s
ro
le
co
ns
ist
s
on
cal
culat
ing
of
dom
inance
relat
ion
s
hip,
and
gen
e
rati
ng
of
s
kyli
ne
ser
vices,
ba
sed
on
the
m
essages r
ecei
ved
fr
om cur
re
nt co
m
po
sit
ion agen
ts. Th
is
operati
on aim
s
t
o
determ
ine the co
m
po
sit
ion
ag
ents
to
be
el
i
m
inate
d
or
to
be
preser
ve
d
f
or
t
he
nex
t
le
vel
com
po
sit
ion
,
or
t
o
determ
ine
the
global
optim
a
l
com
po
sit
ion
.
Co
m
posi
tion
Ag
e
nt:
is
the
com
po
ne
nt
tha
t
con
ta
ins
t
he
QoS
vect
or
of
a
serv
ic
e
or
of
a
par
ti
al
com
po
sit
ion
.
I
ts
ro
le
is
to
se
nd
a
m
essage
con
ta
in
s
the
Q
oS
vecto
r
t
o
t
h
e
c
on
t
ro
l
a
ge
nt,
a
nd
c
on
se
quently
base
d
on
received
decisi
on
from
the
con
tr
ol
le
r,
the
com
posit
ion
te
rm
inate
it
s
execu
ti
on,
or
re
pro
duce
oth
e
r
com
po
sit
ion
a
gen
ts
with
t
he
serv
ic
e
qual
it
ie
s
of
t
he
f
ollow
i
ng
cl
asses,
before
sim
i
lar
ly
com
pleti
ng
it
s
execu
ti
on.
Algorithm
2
descr
ibe
the
dif
f
eren
t
be
ha
vio
r
s
of
the
co
ntr
ol
agen
t
from
init
ia
li
za
ti
on
un
ti
l
the
la
st
ste
p
and
retu
r
ning
the
opti
m
al
res
ult.
It
ta
ke
a
s
input
a
set
of
s
erv
ic
e
cl
asses
of
c
om
po
sit
ion
C=
{c1,
c
2,
c
3…
}
,
wh
e
re
eac
h
e
nt
ry
c
i
de
no
te
s
a
set
of
se
r
vices
that
belo
ngs
to
t
he
i
-
t
h
s
erv
ic
e
cl
ass
.
It
us
e
al
so
an
inte
ge
r
var
ia
ble I
t
o pre
serv
e
the c
urr
ent level
of co
m
po
sit
ion
.
Al
go
rit
hm
2
:
Con
tr
ol
Ag
e
nt
al
gorithm
Inp
ut
:
a s
et
of
serv
ic
e cla
sse
s
of a c
om
po
sit
ion
: C
.
Ou
tp
ut:
optim
al
co
m
bin
at
ions o
f
the
co
m
posit
ion
C:
Co
p
1: nb
_ag
e
nt =
siz
e(C[I
]
)
2: whil
e (i <
nb_
a
ge
nt)
3:
Q.
a
dd(r
ecei
ve())
3:
i+
+
4: end
w
hile
5: Q
op
=
Sk
yl
in
e (
Q)
6: whil
e (i <
nb_
a
ge
nt)
3:
for q in
Q
op
7:
if (q ==
Q
[i]
)
8:
Send
(Q
.
get(i
).getSen
de
r(),
A
GREE)
/*
se
ndin
g an
AGREE
m
e
ssage to
do
minant
agen
ts
*
/
9:
Q.
r
em
ov
e(i)
10:
En
d
if
11
:
end f
or
12:
e
nd
wh
il
e
13
:
f
or
eac
h q i
n Q
Send
(q.g
et
Se
nder
(), CA
NCE
L)
/*
se
ndin
g a C
ANC
EL m
e
ssage to
do
minate
d ag
e
nts
*/
14
: e
nd
for
15
:
retu
rn
Q
op
Algorithm
3
de
fine
the
be
ha
viors
of
the
c
om
po
sit
ion
a
ge
nts,
knowin
g
that
durin
g
e
xecu
ti
on
of
a
com
po
sit
ion
,
s
ever
al
com
po
si
ti
on
age
nts
can
be
gen
e
rated
i
n
f
un
ct
io
n
of
the
num
ber
of
cl
asses,
an
d
nu
m
ber
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
8776
IJECE
V
ol.
6
,
No.
3
,
Decem
ber
20
1
7
:
179
–
188
186
of
se
rv
ic
es
in
each
cl
ass.
O
n
the
con
tra
ry,
on
e
in
sta
nc
e
of
co
ntro
l
a
gen
t
is
gen
erate
d
duri
ng
a
n
exec
ut
ion
.
A
com
po
sit
ion
A
gen
t
ta
ke
as
in
pu
t
a
Q
os
ve
ct
or
of
one
web
serv
ic
e
Q,
or
a
Qos
of
a
par
ti
al
com
po
sit
ion.
The
Qo
s
of
pa
rtia
l
com
po
sit
ion
is
cal
culat
ed
i
n
the
beg
i
nnin
g
of
it
s
li
fe
cy
cl
e,
by
us
in
g
aggre
gati
on
f
unct
ion
betwee
n Qos
ve
ct
or
of it
s cr
e
at
or
a
ge
nt Q_
1, an
d
t
he Qos
of the
ass
ociat
ed
se
rv
ic
e
Q.
Al
go
rit
hm
3
:
Com
po
sit
ion
a
gen
t al
gorithm
Inp
ut
:
Qo
s
v
ec
tor of a se
rv
ic
e
: Q, Q
os
vecto
r
of c
on
tr
ol a
ge
nt: Q
_1, id
e
ntit
y of co
ntr
ol agent:
AgC, set
of
serv
ic
e
cl
asse
s
of the c
om
po
sit
ion
: C
,
t
he
c
urren
t l
e
vel of c
om
po
sit
ion
to
w
hich
t
he
c
orres
pond se
rv
ic
e
belo
ngs: I.
1: if (I>
0)
2:
Q=U(Q
, Q_
1)
//
ca
lc
ula
te
ne
w Qos
attri
bu
t
es o
f
parti
al composit
io
n
wi
th
agg
r
eg
atio
n
f
un
ct
io
ns
U
3: end i
f
4: se
nd(AgC
,
Q)
//
sen
d a
mes
sage wi
th Ve
ct
or
Qo
s
to c
ontrol
ag
e
nt
5: Msg=
receiv
e(AgC)
//
rec
ei
ve con
tr
oller
resp
on
se
6: if (M
sg ==
AG
RE
E)
7:
for
eac
h
S
in
C
[I
+
1]
8:
create
Com
po
sit
ion
A
ge
nt(S.ge
tQualit
e()
,
Q,
I
+1)
//
create a new
composit
io
n
ag
e
nt
9:
end f
or
10
:
sto
p()
//
te
rminate
the e
xe
cution
11
: el
se
if
(Ms
g
==
C
AN
C
EL
)
12
:
stop()
13
: e
nd
If
4.
E
X
PERI
MEN
TAL EV
ALU
ATIO
N
In
t
his
sect
io
n,
we
pr
e
sent
an
e
xperim
e
ntal
evaluati
on
of
our
im
p
rove
d
s
kyli
ne
com
po
sit
io
n
appr
oach,
in
te
rm
s
of
the
qu
antit
y
of
cand
id
at
e
co
m
bin
at
ion
s
to
be
cal
cul
at
ed
and
ge
ne
r
at
ed,
al
so
in
te
rm
s
of
execu
ti
on ti
m
e
r
e
qu
ire
d
t
o rea
ch
the
opti
m
al
so
luti
on.
Our
e
xp
e
rienc
e
is
pe
rfor
m
ed
us
in
g
a
set
of
we
b
ser
vices
gen
e
rated
with
rand
om
qu
al
it
y
at
tribu
te
s
.
These
ser
vices
are
i
de
ntifie
d
by
tw
o
a
bs
trac
t
par
am
et
ers
of
Q
os
t
o
m
ini
m
i
zed,
an
d
w
hich
ta
ke
ra
ndom
values
betwee
n
1
an
d
50.
I
n
t
he
eval
uation,
we
ha
ve
init
ia
li
zed
our
pro
gram
by
a
giv
e
n
c
om
po
s
it
ion
with
15
s
erv
ic
e
cl
asses,
w
her
e
each
cl
ass
co
ntains
20
ge
ner
at
ed
ser
vices,
w
hich
m
eans
that
we
hav
e
a
tot
al
of
15
20
can
di
date
com
bin
at
ion
s
f
or
this
e
xp
e
ri
m
ental
com
po
sit
ion
.
T
he
f
ollow
i
ng
fi
gures
10
an
d
11
repr
esent
the
va
ria
ti
on
of
the
exec
ution
t
i
m
e,
and
of
the
nu
m
ber
of
ge
ner
at
e
d
com
bin
at
ions
as
a
functi
on
of
t
he
num
ber
s
of
t
he
cl
asses
and
ser
vices
of
t
he
c
om
po
sit
io
n.
T
he
di
sp
la
ye
d
values
represe
nts
th
e
ave
rag
e
of
10
exec
utio
ns
of
th
e
al
gorithm
, w
it
h
the
sam
e p
ar
a
m
et
ers
(num
ber
of
classe
s a
nd se
rv
ic
es
),
a
nd
with
diff
e
re
nt
v
al
ues
of se
rvi
ces.
We
i
m
ple
m
ented
the
i
m
pr
ov
ed
sk
yl
ine
co
m
po
sit
ion
al
go
rithm
descr
ib
ed
previ
ou
s
ly
in
Java,
a
nd
us
in
g
J
ADE
pl
at
fo
rm
as
a
m
ulti
-
agen
t
syst
em
,
to
i
m
ple
m
e
nt
our
syst
em
com
po
sit
ion
a
nd
c
ontr
ol
age
nts.
T
he
exp
e
rim
ents w
ere c
onduct
ed on a S
ony l
apt
op m
achine w
i
th an I
ntel co
re
i5 2.
40G
Hz
C
PU
a
nd
8 GB
RAM.
We
m
easur
ed
the
aver
a
ge
execu
ti
on
ti
m
e
and
can
dida
te
co
m
po
sit
ion
gen
e
rated
,
req
ui
red
by
al
gorithm
fo
r
so
lvi
ng
the
gi
ven
ra
ndom
c
om
po
sit
ion
,
de
fining
the
nu
m
ber
of
cand
i
date
serv
ic
es
by
20
serv
ic
es
pe
r
cl
ass,
an
d
15
cl
asses
pa
r
c
om
po
sit
ion.
T
he
re
su
lt
s
of
this
e
xperim
ent
are
presente
d
in
fi
gure
10
and 11.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
ICT
IS
S
N:
22
52
-
8776
An
A
ge
nt Arc
hi
te
ct
ur
e for
Qos
-
B
as
ed
We
b S
ervi
ce Com
pos
it
ion
…
(
El
-
Ala
mi Ay
oub
)
187
Figure
10: M
easur
i
ng the
va
r
ia
ti
on
of
num
ber
of
c
om
po
sit
ion
s
Figure
11: M
easur
i
ng the
va
r
ia
ti
on
of e
xec
ut
ion
ti
m
e fo
r
a
com
po
sit
ion
5.
CONCL
US
I
O
N
In
this
pa
per
,
we
pro
po
se
a
m
et
ho
d
t
o
s
olve
the
pro
blem
of
com
po
sit
io
n
base
d
on
the
qual
it
ie
s
of
the
serv
ic
es
,
us
in
g
the
Sk
yl
ine
m
et
ho
d
an
d
th
e
agen
t
-
or
ie
nte
d
arc
hitec
ture.
The
pro
posed
m
et
ho
d
pe
rfo
rm
by
us
in
g
S
kyli
ne
m
et
ho
d,
wh
ic
h
aim
to
reduce
the
nu
m
ber
of
serv
ic
es
to
be
com
bin
ed
an
d
gen
e
rated,
bas
ed
on
do
m
inati
on
relat
ion
s
hip
betw
een
ca
nd
i
date
s
erv
ic
es.
O
n
t
he
oth
e
r
hand,
to
increase
pe
rform
ance
an
d
s
pe
ed
of
al
gorithm
,
we
decide
t
o
us
e
t
he
par
a
dig
m
of
syst
e
m
agen
ts
to
ta
ke
ad
va
ntage
of
it
s
colle
c
ti
ve
an
d
i
ntell
i
gen
c
e
beh
a
viors
t
o
de
com
po
se
an
d
so
lve
easi
ly
the
c
om
po
sit
ion
prob
le
m
.
To
i
m
ple
m
ent
the
com
po
sit
ion
m
et
hod
we
us
e
a
m
ulti
-
age
nt
syst
em
as
platf
or
m
fo
r
our
pro
gr
a
m
,
wh
ic
h
co
nt
ai
ns
tw
o
ty
pes
of
age
nts,
on
e
for
to
perform
the
com
po
sit
ion
proc
ess,
an
d
ano
t
he
r
on
e
to
c
on
tr
ol
and
the
orc
hestrati
on
com
po
sit
io
n
operat
ion
s
.
Our
e
xp
e
rim
en
ts
hav
e
s
how
n
that
the
perfor
m
ance
of
ou
r
m
et
ho
d
ca
n
be
us
e
fu
l
for
c
om
po
sit
ion
with
a
ve
ry
la
rg
e
qua
ntit
y
of
can
di
date
co
m
bin
at
ion
s,
an
d
w
hich
can
not
be
so
lve
d
with
si
m
ple
exh
aus
ti
ve
search.
Fi
nally
,
us
in
g
a
gen
t
pa
rad
i
gm
and
sky
li
ne
m
et
ho
d
c
an
re
duce
the
e
xecu
ti
on
ti
m
e,
al
so
opti
m
iz
e
t
he
m
e
m
or
y
ta
ken
f
or
storing
data
of
al
l t
hese co
m
bin
at
ions.
REFERE
NCE
S
[1]
M.
Alrifa
i
;
D.
S
kouta
s;
T
.
Riss
e
,
“
Sele
c
ti
ng
Sk
yli
ne
Serv
ic
es
for
QoS
-
base
d
W
eb
Servic
e
Com
p
ositi
on
”
,
Raleig
h
NC USA
,
pp
.
11
-
20,
2010
.
[2]
M.
Alrifa
i
;
T
.
Riss
e.
“
Com
bini
ng
globa
l
op
tim
iz
at
ion
wi
th
l
oca
l
se
le
c
ti
on
f
or
eff
icient
qos
-
awa
re
serv
ic
e
compos
it
ion
”,
In
W
W
W
,
pp.
881
-
890,
2009
.
[3]
D.
Skoutas;
D.
Sacha
rid
is,
“
A.
Sim
it
sis,
and
T.
Sell
is.
Serving
the
sk
y
:
Discov
eri
ng
and
sel
ecti
ng
sem
ant
ic
we
b
servic
es
through
d
y
namic
sk
y
l
in
e
quer
ie
s
”,
In
te
r
nati
onal
Con
fe
re
nce
on
Semant
ic
Computing
,
pp
.
2
22
-
229,
2008
.
[4]
Benoua
re
t;
D.
B
ensl
imane;
A
.
H
adj
a
li
,
“
On
the
Us
e
of
Fuzzy
Do
minance
for
Co
mputing
Serv
i
ce
Sky
l
ine
Based
o
n
QoS
”
.
In
th
e
9th
Inte
rn
at
ion
al Co
nfe
re
nc
e
on
W
e
b
Services (IEE
E
ICW
S 2011).
0
1
0
0
0
2
0
0
0
3
0
0
0
4
0
0
0
5
0
0
0
6
0
0
0
7
0
0
0
8
0
0
0
9
0
0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Number
of g
ener
at
ed
com
bi
na
t
i
on
numbre
of
cla
sses
0
5
0
0
0
1
0
0
0
0
1
5
0
0
0
2
0
0
0
0
2
5
0
0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Ex
ec
ut
i
on
t
i
me
in
ms
numbre
of cl
asses
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
8776
IJECE
V
ol.
6
,
No.
3
,
Decem
ber
20
1
7
:
179
–
188
188
[5]
Yang
Yu;
Jian
Chen;
Shangqu
an
Li
n;
Ying
W
ang,
”A
D
y
n
amic
QoS
-
Aw
ar
e
Logi
sti
cs
Servic
e
Com
positi
on
Algorit
hm
Base
d
on
Social
Net
work”,
IE
EE
transacti
ons
on
e
merging
topi
cs
in
computi
ng
,
v
olume
2,
no
.
4
,
dec
ember
2014.
[6]
Mike
P;
Pap
az
o
glou
W
il
le
m
-
Ja
n
van
den
Heuv
el
,
“
Service
ori
e
nte
d
ar
chi
t
ec
tur
es:
appr
oa
che
s,
te
chno
logi
es
an
d
re
sea
rc
h
issues”
3
M
arc
h
2007
S
pringe
r
-
Verl
ag.
[7]
G.Ca
nfora
,
“
An
approach
for
QoS
-
aware
servic
e
composit
i
on
based
on
gene
tic
algorit
hm
s”,
Proce
ed
ings
of
the
2005
conf
er
ence
on
Gen
et
i
c and evol
uti
on
ar
y
co
m
puta
ti
on
,
pp.
10
69
-
1075
,
2005
.
[8]
Abourez
k;
A.
I
drissi,
“
Introduct
ion
o
f
an
out
ranking
method
in
the
Cloud
c
omputing
research
and
Se
le
c
ti
o
n
Syste
m based
on
the
Sky
l
ine
”,
Re
sea
rc
h
Challeng
es
in
Inform
at
ion
Scie
nce
(RCIS
),
2014
IEE
E
Eighth
In
te
rna
ti
on
al
Confer
ence
on
.
1
-
12,
2014
.
[9]
Paolo
Busetta;
Mass
imo
Za
ncanaro,
“
Open
soc
ial
agen
t
archite
ct
ure
for
distrib
ute
d
mult
imedi
a
.
In
Workshop
on
Age
nts
at
Work:
Deploy
ed
app
li
c
ati
ons
”
,
2nd
Int
ern
ational
Join
t
Confer
ence
on
Autonom
o
us
Ag
ent
s
&
Multi
agent
S
y
stems
(AA
M
AS
2003),
Melb
ourne
,
Vi
ct
or
ia, A
ustral
ia,
Jul
y
2
003.
[10]
Al
-
Masri
E;
Ma
hm
oud
QH
,
“
Qo
S
-
based
discov
ery
and
rank
ing
o
f
web
serv
ices”
,
I
n:
16th
Internat
i
onal
Conf
ere
n
ce
on
Com
pute
r
Co
m
m
unic
at
ions and Net
works
(IC
CCN 2007),
pp
52
9
–
534.
[11]
Börzsön
y
i
S;
K
oss
m
ann
D;
Stocke
r
K,
The
sk
yline
op
erator
.
In:
Proce
ed
ings
of
the
17th
interna
t
iona
l
conf
er
ence
on
data engineer
ing
(ICDE’01)
,
pp
421
–
430
,
201
1.
[12]
S.
W
ang;
Q
Sun;
H.
Zou;
F.Yan
g,
“
Parti
c
le
Sw
a
rm
Optimiza
ti
on
with
Sk
y
l
ine
O
per
at
or
fo
r
Fa
st
Clou
d
-
base
d
W
e
b
Servic
e
Com
positi
on”
,
Mob
ile
N
et
w
Appl
,
pp.
11
6
–
121
,
2013
.
[13]
T.
Yu;
Y.
Zha
ng
;
K.
-
J.
L
in,
“
Eff
i
ci
en
t
al
go
rit
hm
s
for
web
services
sele
c
ti
on
wi
th
e
nd
-
to
-
end
qos
co
nstrai
nts”
,
AC
M
Tr
ans.
on
the W
eb
,
1(1
), 2007.
[14]
M.
Abourez
q
;
A
.
Idrissi,
“
Introd
uct
ion
of
an
out
ra
nking
m
et
hod
in
the
Cloud
Co
m
puti
ng
Resea
r
ch
and
Sel
ec
t
ion
S
y
stem ba
sed
on
the Sk
y
li
ne
,
”
,
28
-
30
Ma
y
2014
.
[15]
Kang;
K.
M
.
Sim
,
"
Cloudl
e
A
n
Age
n
t
-
based
Cloud
Searc
h
Engi
ne
that
Co
nsults
a
Cloud
Ontology
",
C
lou
d
Com
puti
ng
and Virtua
liza
ti
on
C
onfe
re
nc
e, 2010.
[16]
A.
Idrissi;
M.
Abourez
q,
"S
k
y
li
ne
in
Cloud
Com
puti
ng",
Journal
of
Theoretical
and
Applied
Informatio
n
Technol
ogy
,
Vol
.
60
,
No.
3,
Febr
uar
y
2014.
[17]
C.
Ze
ng,
X.
Guo,
W
.
Ou
and
D.
Han,
Cloud
Com
puti
ng
Servic
e
Com
positi
on
and
Sear
ch
Based
on
Sem
ant
ic
,
Cloud
Computin
g,
Vol
.
5931
,
pp
.
290
-
300
,
2009
.
Evaluation Warning : The document was created with Spire.PDF for Python.