TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.7, July 201
4, pp
. 5559 ~ 55
6
6
DOI: 10.115
9
1
/telkomni
ka.
v
12i7.464
3
5559
Re
cei
v
ed O
c
t
ober 7, 20
13;
Revi
se
d Ja
n
uary 7, 2014;
Acce
pted Fe
brua
ry 2, 201
4
Intuitionistic Fuzzy Set Based Task S
c
heduling with
QoS Preference Awareness
Wang J
u
an
Schoo
l of Net
w
ork Engi
ne
erin
g, Chen
gd
u
Un
iversit
y
of Information T
e
chno
log
y
Chen
gd
u, Chi
n
a
email: w
a
ng
ju
a
n
@cuit.e
du.cn
A
b
st
r
a
ct
Existing task s
c
hed
uli
ng
alg
o
r
i
thms
in c
l
ou
d
comput
i
ng h
a
v
e
lack th
e a
b
il
ity to be aw
ar
e
of users'
qua
lity of serv
ice (QoS) pref
erenc
e.
In ord
e
r to addr
ess
this prob
le
m, "
i
ntuiti
onistic fu
zz
y
a
nalys
is" i
s
introd
uced
to
d
e
termin
e
us
ers
'
QoS prefer
en
ce. In a
d
d
i
tion,
the
"opti
m
al s
equ
enc
e d
e
cisi
on
metho
d
"
hel
p
s
experts
use t
h
eir pr
ofessi
on
a
l
kn
ow
ledg
e to
deci
de th
e weights of QoS
classes.
Us
ing
these
metho
d
s,
w
e
prop
ose the "I
ntuitio
n
istic F
u
zz
y
Set Based
T
a
sk
Scheduli
ng w
i
th QoS Preferenc
e Aw
areness: IF
S-Qo
S
PA" alg
o
rith
m.
By consid
erin
g both us
er a
nd ex
per
t exp
e
rie
n
ce, the
method c
an d
e
termine
users'
QoS
preference and reflect the c
haracter
i
stic of
cloud
storage system
. The
simulation
r
e
s
u
lts show that
this
meth
od offers
accepta
b
l
e
us
er satisfaction
rate, has
low
computati
on c
o
mpl
e
xity an
d mor
e
suita
b
le
for
larg
e scale tas
k
schedu
lin
g a
s
comp
are to p
a
rticle sw
arm
o
p
timi
z
a
ti
on bas
ed on
es.
Ke
y
w
ords
:
qu
ality of service,
preferenc
e aw
aren
ess, intu
iti
onistic fu
zz
y
se
t, task scheduli
ng, clou
d stora
g
e
Copy
right
©
201
4 In
stitu
t
e
o
f
Ad
van
ced
En
g
i
n
eerin
g an
d
Scien
ce. All righ
ts reser
ved
.
1. Introduc
tion
With the dev
elopme
n
t of the Intern
et of Thi
ngs
(IoT
), the Interne
t
extends to
variou
s
terminal
s an
d create
s
ex
plosive g
r
o
w
t
h
data.
Fo
r the hu
ge am
ounts
data,
not only are
the
hard
w
a
r
e a
n
d
softwa
r
e
s
i
n
vestment b
e
y
ond imagin
a
t
ion,
but also
the lack of professio
nal d
a
ta
maintena
nce
admini
s
trato
r
make
it i
s
im
possibl
e
for
i
ndividual
s
to
build data ce
nter
th
emselves.
In ord
e
r to d
eal
with the
huge
amo
unt
s of
data, “cloud storage
”
has
be
en su
gge
sted as a
solutio
n
for these pro
b
le
ms. The idea
of cloud
sto
r
age i
s
that the profe
s
sio
nal clou
d sto
r
age
servi
c
e
r
build
the clou
d sto
r
age
system i
n
clu
de
ha
rd
ware a
nd software a
nd p
r
ovide the “stora
ge
servi
c
e
”
to in
dividual u
s
ers acco
rdi
ng to their
ne
ed.
This way avoids the
rep
e
a
t investment
in
both hardware and software and save t
he co
stly maintenan
ce
cha
r
g
e
.
Clou
d stora
g
e
sy
stem dea
ls with
h
uge
amount
s data
and
tasks.
The global
th
roug
hput
improvem
ent, resource
op
timization
an
d p
r
ofit
maxi
mization
a
r
e the
ultimate
obje
c
tives of the
system [1
-14]
. Task sch
e
d
u
ling al
gorith
m
s a
r
e d
e
si
g
ed to ad
dre
ss these
proble
m
s a
nd pl
ay an
importa
nt rol
e
. There a
r
e
al
ready
many t
a
sk
sched
ulin
g alg
o
rithm
s
i
n
cl
oud
comp
uting a
r
ea,
b
u
t
few o
n
e
s
in
cl
oud
storage
system. The
s
e
schem
es
ai
m to get
high
er
system
through
put, nam
ely
sho
r
ter ma
ke
spa
n
(
the tim
e
differe
nce
betwe
en the
start a
nd fini
sh
of a
seq
u
ence of jo
bs or
tasks), su
ch
as the Min
-
Min and M
a
x-Min alg
o
rith
m [1] which
are e
nume
r
a
t
ion method
and
enume
r
ate all
the possible
solutio
n
s a
n
d
sele
ct one
of
it as the opti
m
al sol
u
tion. Whe
n
num
be
r
of instan
ce
s i
s
la
rge, the
s
e
method
s
co
st unac
ce
pted
time and
sp
a
c
e
whi
c
h m
a
ke them a
r
e
n
o
t
feasibl
e
for sche
duling, th
en the heu
ristic algor
ith
m
is su
gge
sted
to find reaso
nably solutio
n
s,
su
ch as a
n
t colony based
sched
uling al
gorith
mm [2], genetic alg
o
r
ithm (GA) b
a
se
d sched
ul
ing
algorith
m
s [3
, 4], simulated Anne
aling
(SA)ba
se
d sched
uling al
gorithm
s [5], particl
e swa
r
m
optimizatio
n(PSO) [6]. An
other
aim of
sched
uling
al
gorithm
s i
s
lo
ad b
a
lan
c
ing
(ma
k
e ta
sks
are
disp
atchi
ng t
o
re
sou
r
ce n
ode
s
average
ly), includ
e weighted m
ean
time(WMT)
algorithm [7] and
some of he
uri
s
tic alg
o
rithm
s
[7-10].
In re
cent
yea
r
s, th
e q
uality of servi
c
e
(Q
oS) ha
s re
ceived
in
crea
sing
attentio
n whi
c
h
i
s
use
d
to quan
titatively mea
s
ure aspe
cts
of the net
work se
rvice, su
ch a
s
error rates, band
wi
dth,
throughput, transmi
ssi
on
delay, av
ailability, jitter, etc. M
any QoS guided task-scheduling
method
s hav
e been p
r
o
p
o
s
ed, such a
s
QoS Guide
d
Min-Min h
e
u
r
istic [1
1], based o
n
the M
i
n
-
Min algorith
m
, which con
s
i
ders network band
widt
h a
s
the QoS parameter. The
heuri
s
tic met
hod
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5559 – 55
66
5560
whi
c
h integ
r
a
t
e the QoS Min-Min h
e
u
r
i
s
tic an
d
the WMT he
uri
s
tic (Q
WMTM
)
[12] can not
only
guarantee Q
o
S but also g
uara
n
tee loa
d
balan
ce
. F
e
w multi-QoS
guara
n
tee al
gorithm
s [13-15]
use multipl
e
workflo
w
s o
r
multiple co
m
pone
nts to se
parate d
eal with single Q
o
S.
These met
h
o
d
s
are
sim
p
l
y
dra
w
n from
clou
d
comp
uting an
d la
ck of Q
o
S p
r
e
f
eren
ce
awareness
(P
A) ability, whi
c
h i
s
im
portant
for users. Furt
hermore,
most
of these algorithm
s
take
the system th
roug
hput imp
r
oveme
n
t as
the goal
a
nd
ignore the u
s
er’s
req
u
irem
ents an
d re
sult
in low
user satisfa
c
tion
rate. Fin
a
ll
y, the
QoS factors a
r
e
too professional to u
s
e
r
s t
o
unde
rsta
nd.
So the
wei
g
h
t
s of
QoS fa
ctors are u
s
u
a
lly deci
ded
b
y
re
sea
r
che
r
s o
r
te
ch
nici
a
n
s
and do n
o
t re
flect the use
r
'
s
real d
e
ma
n
d
.
In ord
e
r to
a
ddre
s
s a
bove problem
s,
we tran
sfer t
he techni
cal
QoS facto
r
s i
n
to user
unde
rsta
nda
b
l
e QoS
cla
sses, a
nd int
r
o
duce the
“int
uitionisti
c
fu
zzy an
alysis (IFS)” into
ta
sk
sched
uling a
r
ea to help u
s
ers
de
scribe t
he impo
rtan
ce of QoS c
l
as
ses
.
By this
w
a
y ever
y QoS
cla
s
s get it
s wei
ght fro
m
users’ intuiti
on a
nd
refle
c
t u
s
ers’
re
q
u
irem
ent. Th
e propo
se
d t
a
sk
sched
uling i
s
judgedi
ng b
y
the “intuitionistic fu
zzy comprehe
nsiv
e
evaluation value
(IFCEV
)”.
The hi
ghest
IFCEV node will
dispatch to task
and update the IFCEV for
next scheduli
ng.
Simulation
re
sults sho
w
t
hat comp
are
d
with
exi
s
ting alg
o
rithm
s
,
the
pro
p
o
s
e
d
alg
o
rithm
can
satisfy user'
s
QoS prefe
r
en
ce, and h
a
s l
o
w complexit
y
and high ex
ecutio
n efficie
n
cy.
The re
maind
e
r of this m
anu
script is
orga
nized a
s
follows. Se
ction 2 descri
bes the
details
of our method that
introdu
ce
s i
n
tuitionist
i
c
fuzzy analy
s
is into task scheduli
ng in
cl
oud
stora
ge to sa
tisfy QoS pre
f
eren
ce. The
simulati
o
n
s
a
nd co
mpa
r
iso
n
s an
alysi
s
a
r
e presented
in
Section 3. Fin
a
lly, Section 4 pre
s
ent
s a sho
r
t con
c
lu
si
on and future
works.
2. IFS Analy
s
is Bas
e
d Ta
sk Scheduli
ng
w
i
th QoS
Prefer
ence
A
w
a
r
ene
ss (IFS-QoS PA)
The rea
s
o
n
s
why existing
sched
uling al
gorithm
s lack suppo
rt for QoS PA and do not let
use
r
s the
m
se
lves to deci
d
e
the priority le
vel maybe the followin
g
:
(1)
There a
r
e
co
ntradi
ction
s
a
m
ong
QoS fa
ctors,
such a
s
tra
n
smi
s
sio
n
sp
eed
and
transmissio
n
quality. In ord
e
r to e
n
sure
the tra
n
smi
s
si
on qu
ality, the timeout d
e
tection
and
ret
r
an
smi
ssi
on
mech
ani
sm
a
r
e
nee
d
whi
c
h cost
mo
re
time, namely
ma
ke th
e transmi
ssion
speed
sl
owly
down. And the co
st is also
has
contra
di
ction with
sp
eed an
d
qual
ity. Lower co
st and hig
her
spe
ed are ne
ed by use
r
. Ho
wever
syst
em can
not
u
s
ually satisfy both of them at the sam
e
time.
(2)
User l
a
ck
of
the kno
w
ledg
e of te
chni
ca
l
QoS fa
ctors that m
a
ke
use
r
s cann
ot de
cide th
e
importa
nt level of them.
In orde
r to ov
ercome the
a
bove problem
s, we int
r
odu
ce IFS into ta
sk
sched
uling
to help
use
r
s d
e
scrib
e the importa
nce of QoS
cl
asse
s.
In this secti
on, the relat
ed definition
s
are given firstly, and then the task
scheduli
ng
based on IFS
is described i
n
detail.
2.1. Intuitionistic Fuzz
y
Set Rela
ted
D
e
finitions
IFS [16, 17] i
s
exten
ded
from fu
zzy set t
heo
ry, whi
c
h u
s
e
s
the
d
egre
e
of
me
mbershi
p
and
deg
ree
o
f
non-memb
e
r
shi
p
to
de
scribe the
un
ce
rt
ain informati
on, an
d for th
is a
d
vantage
it
is used in ma
ny fields alre
ady.
Defini
tion 1. IFS:
Let a set
X
fixed, then
A
={(
x
,
μ
A
(
x
),
ν
A
(
x
))|
x
∈
X
}is a intui
t
ionistic
f
u
zzy
set
,
wh
ere
μ
A
(
x
) i
s
th
e de
gre
e
of
membe
r
ship
(
the pe
rcent o
f
x
be
lo
ng
to
A
) an
d the
ν
A
(
x
)
is the d
e
g
r
e
e
of non
me
mbershi
p
(the
perce
nt of
x
not belo
ng t
o
A
) of the
element
x
∈
X
to
A
∈
E
,respe
ctively. The functio
n
s
μ
A
and
ν
A
sh
ould sat
i
sf
y
con
d
it
ion:
0
≤
μ
A
(
x
)+
ν
A
(
x
)
≤
1,
x
∈
X
,
μ
A
∶
X
→
[0,1],
ν
A
:
X
→
[0,
1
]
。
Defini
tion 2.
Indetermin
ac
y
Degree
(
π
):
For any
IFS
A
, Let
π
A
=1
-
μ
A
(
x
)-
ν
A
(
x
),
x
∈
X
denote
s
the
degree of ind
e
termin
acy (I
ntuitionisti
c
Index)
which
mean
s the u
n
ce
rtainty level of
x
∈
A
, wh
ere 0
≤
π
A
(
x
)
≤
1,
x
∈
X
。
Defini
tion
3.
Intuitionistic Fuzz
y
Number (
IFN
):
The triple [
μ
(
x
),
ν
(
x
),
π
A
] i
s
def
ined a
s
the intuitionistic fuzzy num
ber(
IFN
). The
n
the set of
IFN
s is also d
e
fined a
s
the
IFS
, denote as
A=
{[
μ
(x
)
,
ν
(x
)
,
π
A
]|
x
∈
X
}
。
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Intuitionistic F
u
zzy Set Based Task Sche
duli
ng
with Q
o
S Prefere
n
ce Awarene
ss (Wang
Juan
)
5561
Defini
tion 4. Intuitionistic
Fuzzy
Weight (
IFW
):
Let the wei
g
ht coefficie
n
t
w
rep
r
e
s
e
n
ts the rel
a
tive importa
nce
level for on
e QoS
attribute to
all QoS
attributes. T
h
is
rep
r
e
s
ent
s t
he attrib
ute’s impa
ct on
comp
re
hen
si
ve
evaluation
when
othe
r Q
o
S attribute
s
a
r
e fixed.
Usu
a
lly, use
r
’
s
Q
o
S re
qui
reme
nts p
r
efe
r
en
ce is
descri
bed by
langu
age, su
ch a
s
import
ant, not im
portant, and ord
i
nary. He
re we use lan
gua
ge
descri
p
tion to
get the weig
ht by Table 1 definition.
In Table 1, th
e un
kno
w
n v
a
riabl
e
π
rep
r
ese
n
ts ind
e
te
rmina
c
y deg
ree of user to
the QoS
attribute. And
different use
r
may give the diffe
rent ind
e
termin
acy d
egre
e
. We u
s
e percenta
g
e
to
descri
be the i
ndetermina
cy
degre
e
π
,
π
∈
(0~100%
).
Table 1. The
Lang
uage
De
scription of
QoS Req
u
ire
m
ents Preference [16]
Lang
ua
ge des
c
riptio
n
Intui
t
io
nistic
fu
z
z
y
nu
mb
ers
Ver
y
impo
rtant
（
5
）
[0.9,0.1,
]
Important
（
4
）
[0.7,0.3,
]
Ordina
r
y
（
3
）
[0.5,0.5,
]
Not important
（
2
）
[0.3,0.7,
]
Not important a
t
all
（
1
）
[0.1,0.9,
]
Then the
IFW
for QoS attribute is defin
e
d
as:
w
j
=
μ
(
q
j
)-
ν
(
q
j
)×
(3)
Whe
r
e
w
j
∈
[0,
1
],
∑
w
j
=1,
j
=1,
2
,…, n. So co
nversi
on
pro
c
ess is nee
d t
o
ma
ke
weig
ht
w
j
fall
in [0, 1]. Convers
i
on formula is
:
w
j
=
w
j
/(
w
1
+
w
2
+…
+
w
n
) (j=1,2,…,n)
(4)
Defini
tion 5. Intuitionistic
Fuzzy
Comprehen
s
iv
e
Ev
aluation Value (
IFCEV
):
A
ssu
me
x
1
,
x
2
,…,
x
n
i
s
a group
Qo
S attribute
s
(cla
sses) of
can
d
idate
no
des,
and
w
1
,
w
2
,…,
w
n
(
w
j
∈
[0,1],
∑
w
j
=1,
j
=1,2,…,n ) are
IF
W
for the
s
e
ca
ndid
a
te
nod
es.
The
n
the
weig
hted me
an sum i
s
def
ined a
s
the
IFCEV
as:
IFCEV
=
x
1
w
1
+
x
2
w
2
+…+
x
n
w
n
=
∑
x
j
w
j
(5)
IFCEV
is a co
mpre
hen
sive
evaluation for candi
date no
de to ce
rtain QoS cla
s
s task. Th
e
can
d
idate n
o
de with big
g
e
r
IFCEV
can
satisfy user Q
o
S requi
rem
e
nt well than the lower
IFCEV
one.
2.2. The TQC Transfer to I
Q
C b
y
Optimal Sequence
Metho
d
(OS
M
)
As mention
e
d
above, the use
r
s la
ck of the
kno
w
led
ge to deci
de
the importa
nt level of
t
e
chni
cal
Qo
S
f
a
ct
or.
Thi
s
is w
h
y
re
sea
r
ch
er
s d
e
ci
de
d t
o
st
op l
e
t
t
i
ng u
s
er
s t
h
e
m
selv
e
s
t
o
d
e
cid
e
QoS factor
weig
ht. We want to find a balanc
e betwe
en profession
al experien
c
e a
nd use
r
intuitive expe
rien
ce
by
cla
ssifying
the
s
e techni
cal
Q
o
S facto
r
s (T
QF) into
user und
erstan
da
ble
intuitive QoS cla
s
ses (I
QC), and implem
ent IFS amon
g IQC.
Acco
rdi
ng to
the existin
g
rese
arch
and
exper
ie
nce [1
-14], the
r
e
are three
main
cla
s
ses
that users ta
ke care
of. Th
ey are
co
st, time an
d
qualit
y. By this wa
y, use
r
s can
only weight t
he
cla
s
ses by th
eir intuitive
e
x
perien
c
e
in
stead of fa
ce
to many te
ch
nical
QoS fa
ctors. L
e
t u
s
e
r
to
descri
be the
importa
nce
of QoS factor that
they
totally do
not und
ersta
nd is
extre
m
ely
unre
a
sona
ble
.
In these th
re
e main IQ
C,
there
are
ma
ny te
chni
cal
QoS
f
a
ct
o
r
s.
We t
r
an
sf
er t
e
ch
nic
a
l
QoS factors i
n
to intuitive class by usin
g OSM as follo
ws:
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5559 – 55
66
5562
Assu
me the
r
e are
n
fac
t
ors (
c
i
) i
n
IQ
C an
d the
s
e
factors h
a
ve
been
no
rma
lized
as
se
ction 3 d
e
scrib
ed. Let
w
i
be the weig
ht of the fact
or(
c
i
),
sat
i
sf
y
[0
,
1
]
i
w
，
1
i
i
w
. The
weig
ht of class
c
i
is
defined as
:
12
12
...
n
i
n
ci
i
i
W
w
inde
x
w
inde
x
w
inde
x
(
6)
The
w
i
is d
e
ci
ded by OSM to redu
ce
subj
ective judgm
ent.
Firstly, dete
r
mine
scale i
s
de
scribe
d by
5 level
s
,
with
high
er l
e
vels indi
cating
th
e hig
her
importa
nce. Then
comp
are factors
co
u
p
le by co
uple
,
if one factor’
s
impo
rtan
ce
level is set to 5,
then the an
other o
ne’
s im
portan
c
e l
e
ve
l is 0; if one
is 3, then an
other is
2. By this way, the
judgem
ent m
a
trix is built which i
s
a
n
n
square matrix,
n
is the nu
mb
er of facto
r
s.
And the value
w
ij
(ro
w
i
, c
o
lumn
j
) is the importa
nce of factor
i
comp
are to
j
,
such
as
w
ij
=3, then
opposite
w
ji
=5-
3=2
whi
c
h in
dicate the im
portan
c
e
of factor
j
comp
are to
i
,
i
j
. The sum of rows
i
i
w
indicate
the impo
rtan
ce of fa
ctor
i
in all facto
r
s. Take the
su
m of all rows and
colu
mn
s
ij
ji
w
as th
e
denomi
nato
r
whi
c
h indi
cat
e
the impo
rtance level of
factor
i
in al
l factors, and
the
i
i
w
as
nume
r
ator, th
en the quotie
nt is the weig
ht of the factor
i
:
()
/
(
)
ii
i
j
i
ji
Ww
w
(7)
The hierarch
ical wei
ghts
bala
n
ce
p
r
ofessi
on
al a
nd u
s
e
r
i
n
tu
itive experie
nce
s
by
con
s
id
erin
g b
o
th of them as sh
own in Figure 1.
Figure 1. Balanci
ng Profe
s
sion
al and
User Intu
itive Experie
nces u
s
i
ng Hie
r
a
r
chical Weig
hted
IFS Analys
is
2.3. IFS Based Task Sch
e
duling
w
i
th
QoS Prefer
e
n
ce A
w
a
r
en
ess (IFS-Qo
S
PA)
Based
on the
above definit
ions a
nd met
hod
s, we p
r
o
posed the IF
S-QoS PA al
gorithm
as sho
w
n in
Figure 2.
Tasks
=
{
t
1
,
t
2
,…,
t
n
}
is the
task
matrix which
waitin
g for di
spat
chin
g in unit time
; Every
t
a
sk
t
i
is
a task
vec
t
or that has
tas
k
properties
,
t
i
=[
q
1
,
q
2
,
…
,
q
n
];
Links is the reso
urce no
d
e
s matrix. if
i
≠
j
, then the entry
L
ij
is a link vector betwe
en
node
i
an
d n
ode
j
th
at ha
ve link
pro
p
e
r
ties;if
i
=
j
, then the
L
ij
actu
ally indicate t
he p
r
op
ertie
s
of
node
i
;
Sched
ule Ve
ctor=
[
v
1
,
v
2
,…
.
,
v
n
]
is the
task
sche
du
ling vecto
r
, namely a
scheduli
n
g
solutio
n
. In cloud
storage
system,
v
i
re
pre
s
ent th
e
data of
i
-
th
task
is offer
by the
v
i
n
u
m
ber
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Intuitionistic F
u
zzy Set Based Task Sche
duli
ng
with Q
o
S Prefere
n
ce Awarene
ss (Wang
Juan
)
5563
node.
The
n
t
he le
ngth
of
V
is th
e
sum
of tasks
whi
c
h waiting fo
r
disp
atchi
ng i
n
unit tim
e
. T
here
is an exam
ple: one task sche
duling ve
ctor is [4, 1, 3,
1, 2, 6, 5]. The length
of vector i
s
7, so
the
task n
u
mbe
r
is 7 that mea
n
s in unit time there a
r
e 7
tasks nee
de
d to sch
edul
e. The value of
seq
uen
ce 1 i
s
4 which me
an task 1’
s d
a
ta is offered
by node 4. S
i
milarly, node
1 offers d
a
ta
to
task 2 an
d 4; node 3 offers data to task 3; node 2 offers data to
task 5; node
6 offers data
to
tas
k
6; node
5 offers
data to tas
k
7.
Figure 2. The
Flow Ch
art o
f
the IFS-QoS PA Algorithm
3. Simulation and Analy
s
is
We
develo
p
e
d
a
Clo
ud St
orag
e Simul
a
tion System
(CS
3
) u
s
in
g M
a
tlab7.0.
Thi
s
system
inclu
d
e
s
thre
e main mod
u
l
e
s: task sche
duling mo
dul
e, update an
d
the evaluatio
n module.
The row
of task matrix i
s
the task ve
ctor
whi
c
h
contain
s
task
size an
d task’ QoS
requi
rem
ents,
den
oted
as task
(
T
si
z
e
,
q
1
,
q
2
,…,
q
n
).The orde
r of
QoS fa
ctor
as th
e Se
ction
2
defined. If ta
sk do
es n
o
t require ce
rtai
n QoS fa
ctor,
the acco
rdin
g value set to be NULL. The
node
s matrix
contai
ns the i
n
formatio
n of node
s, su
ch
as the no
de
QoS simila
r to task ve
ctor
and
the nod
es rel
a
tion de
scri
b
e
we
athe
r two nod
es
are
con
n
e
c
ted o
r
not, den
oted
as
nod
e vector
(
nod
e
i
-
n
ode
j
,
q
1
,
q
2
,…,
q
n
),where if
i
=
j
, the
q
describ
e
the
node
i
’ Q
o
S factor, an
d if
i
≠
j
, t
h
e
q
descri
be the
con
n
e
c
t QoS factor b
e
twee
n nodei a
nd n
odej, namely
the links.
The syste
m
take
s the task and node m
a
trixes
a
s
the
input. Then the “task sch
e
duling
module
”
di
sp
atche
s
the
ta
sks. Th
e up
d
a
te mod
u
le u
pdate
s
the
Q
o
S facto
r
val
ue of the
nod
es
matrix after one sch
eduli
ng sche
me i
s
appli
ed.
Fi
nally, the evaluation mo
d
u
le evaluate
s
the
s
c
heme effect by us
er satis
f
ac
tion rate.
If a task i
s
di
spat
che
d
to a
node, the
n
we com
p
a
r
e th
e task an
d no
de vecto
r
s,
when all
the QoS factors of the ta
sk a
r
e satisfi
ed, the
sch
e
duling sch
e
m
e
is con
s
id
ered to satisfy the
tas
k
.
Us
er satis
f
ac
tion rate
is
defined as
:
(%)
%
s
a
tis
fie
d
t
a
s
k
n
u
m
b
e
r
US
R
to
ta
ll
ta
s
k
n
u
m
b
e
r
(8)
In existing al
gorithm, o
n
ly a few he
uri
s
tic algo
rithm
s
, such a
s
PSO and
GA, can offer
multi-QoS
constrai
nt abilit
y by redefined the fitn
ess function [13,
14]. Thei
r fi
tness functions
inclu
de m
any
QoS fa
ctors with
differe
n
t
we
ig
hts. T
h
e US
R
com
p
arison
betwe
en PSO
ba
se
d
algorith
m
and
our IFS base
d
algorith
m
are sho
w
n in Fi
gure 3.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5559 – 55
66
5564
We
ca
n
se
e t
he
USR of P
S
O ba
se
d al
gorithm
is ob
viously lo
we
r than
the
US
R of
ou
r
IFS-QoS PA algorith
m
. Th
e rea
s
o
n
is t
he wei
ghts
u
s
ed in PSO’
s
fitness fun
c
ti
on are fixed, but
the user’
s
PA
is
cha
nging.
If the weig
hts of
PSO’s fitn
ess fun
c
tion
are
ch
angin
g
with u
s
e
r
’sP
A
,
doe
s the PSO based alg
o
r
ithm offer satisfacto
ry effici
ency an
d US
R?
Figure 3. The
USR Co
mpa
r
sio
n
betwee
n
the PSO and IFS-QoS PA Algorithms
We u
s
e IQC
(incl
ude thei
r weights) to redefin
e
d
the fitness fun
c
ti
on of PSO to
make it
reflect the
user’s PA. And
high level ta
sks
are
di
spat
che
d
firstly, then the lo
we
r one
s.Th
e le
vel
of tasks is d
e
f
ined as the
highe
st level 1 is the
TQ t
a
sks, level 2
is Q tasks, level 3 is T tasks
and finally the lowe
st level 4 is the C tasks. Thi
s
improved by intuitionisti
c
fuzzy analysi
s
PSO
is
calle
d IFS-P
S
O, and it
s solution
spa
c
e
is limited
by
existing m
a
tri
x
[6]. The US
R
comp
ari
s
o
n
of
the imp
r
oved
IFS-PSO an
d IFS-Q
o
S PA is
s
h
ow
n as
F
i
gu
re
4
.
In o
r
de
r to
re
mo
ve
u
n
e
x
p
e
c
te
d
interferen
ce,
we repe
at the simul
a
tion
10 time
s
an
d get their
mean valu
es. The simul
a
tion
results
sh
ow
the US
R of th
ese
two
algo
rithms
a
r
e
alm
o
st the
sa
me.
Sometime
s t
he
USR
of the
IFS-PSO is
lower than IFS-QoS PA is
bec
a
us
e the PS
O fall into the loc
a
l optimal.
Figure 4. Co
mpari
s
o
n
of the Improve
d
IFS-PSO and
IFS-QoS PA
Ho
wever, the
IFS PSO cost obviously m
o
re time tha
n
the IFS-QoS
PA which m
e
ans the
PSO has lo
wer effici
en
cy. As shown
in the Fi
gure 5, whe
n
the task scal
e increa
se, the
executio
n time of IFS-PSO is in
cre
a
se sha
r
pl
y. Th
e executio
n time of IFS-Q
o
S PA is linear
gro
w
th and m
o
re suitable f
o
r large scal
e
task.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Intuitionistic F
u
zzy Set Based Task Sche
duli
ng
with Q
o
S Prefere
n
ce Awarene
ss (Wang
Juan
)
5565
Figure 5. The
Execution Ti
me of Multi-Q
o
S IFS and IFS-PSO
4. Conclusio
n
In this p
ape
r, we
studi
e
d
sche
dulin
g
al
go
rithms for
cl
oud sy
stem
s.
The existing
algorith
m
s l
a
ck th
e a
b
ility to sup
p
o
r
t m
u
ltiple QoS fa
ctors a
nd PA. IFS is introd
ued to
hlep
u
s
er
desribe thei
r QoS requirement, and the “IFS-Q
oS PA”sch
eduli
ng algo
rithm
is propo
se
d
.
In
addition, th
e
OSM i
s
u
s
ed to
help
e
x
perts
use
t
heir professi
onal kn
owle
d
ge
to de
cide
the
weig
hts of Q
o
S cla
s
ses.
By consi
deri
ng both
u
s
e
r
and expe
rt experie
nce, the metho
d
can
determi
ne u
s
ers' PA and
reflect the
ch
ara
c
teri
st
ic o
f
cloud sto
r
a
ge syste
m
. The sim
u
lation
s
sho
w
thi
s
me
thod n
o
t only
offers multipl
e
QoS
con
s
traint an
d
satisfies u
s
e
r
’s Q
o
S PA, but al
so
has lo
w comp
utation com
p
l
e
xity and is su
itable for la
rge scale ta
sks sche
dulin
g.
Duri
ng the
si
mulation, we
found
whe
n
the resour
ce
n
ode
s a
r
e fixe
d, the task di
stributio
n
is
influ
e
n
c
e
t
he USR. The
rel
a
tion
ship
betwe
en
ta
sk
dist
ributio
n and US
R
p
r
o
v
ides a guide
to
cho
o
se a fit task
distri
buti
on for the
sy
stem
when t
he re
so
urce
node
s
cann
o
t
be increa
se
d
immediately.
Ho
wever, th
e
rule
s
betwe
e
n
the ta
sk di
stribution
and t
he US
R
need
more research
in the future.
Ackn
o
w
l
e
dg
ement
This work wa
s su
ppo
rted i
n
part by gra
n
ts from:
1.
The Appli
c
ati
on Basi
c Re
search Pr
oj
ect
of Sichuan P
r
ovince (2
013
JY006
4);
2.
The Sci
entific Resea
r
ch
Foun
dation
of
CUIT (Ch
eng
du University
of
Information
Tech
nolo
g
y), No.KYTZ20
1
121);
3.
The Re
se
arch Fund You
n
g
of Middle-
A
ged Aca
demi
c
Lea
ders in
CUIT
(
J201
10
7);
4.
The Sich
uan
Province Scie
nce a
nd Te
ch
nology Supp
o
r
t Plan(No.20
11GZ0
195
);
5.
The Scie
ntific Re
se
arch
Project of
E
ducation
Depa
rtmen
t
in Sichua
n Province
(
N
o
.
10
Z
B
0
93)
.
Referen
ces
[1]
Sandeep T
a
y
a
l. T
a
sks Sc
hed
uling optimization
for the Cloud Com
puting S
y
stems.
(IJAEST)
Internatio
na
l Journ
a
l Of Adva
nced En
gi
neer
i
ng Scie
nces A
nd T
e
chn
o
l
ogi
es
. 2011; 5(
2): 111-
115.
[2]
Xi
an
gqi
an S
o
n
g
, Lin G
ao, Ji
e
p
in
g W
a
n
g
. Jo
b sche
d
u
lin
g b
a
sed
on
ant c
o
lo
n
y
o
p
timizat
i
on
in cl
ou
d
computi
ng.
IEEE.
2011.
[3]
Che
nho
ng Zh
ao, Sha
n
sh
an
Zhang, Qi
ng
feng L
i
u, Ji
a
n
X
i
e, Jic
h
e
n
g
Hu. In
dep
e
nde
nt T
a
sks
Sched
uli
ng Ba
sed on Ge
netic
Algorithm i
n
Cl
oud C
o
mputi
n
g
. IEEE.
2009.
[4]
Yuji
a Ge, Gui
y
i
W
e
i. GA-Based T
a
sk Schedu
ler for the Clo
u
d
Comp
uting S
y
stems
. IEEE.
201
0.
[5]
Gan Guo-
nin
g
, Hua
n
g
T
i
ng-le
i,
Gao Sh
ua
i.
Genetic sim
u
l
a
ted a
nne
al
ing
alg
o
rithm for t
a
sk sche
dul
ing
base
d
on cl
oud
computin
g env
ironme
n
t.
IEEE.
2010.
[6]
W
ang ju
an, Li
F
e
i, Z
hang l
uqi
ao. T
a
sk Schedul
e Al
g
o
rithm
in Cl
oud Stor
a
ge S
y
stem usi
ng PSO
w
i
t
h
Limited Solution Domain.
App
licatio
n Res
ear
ch of Co
mpute
r
s
. 2013; 30(
1): 127-1
29, 15
4 (In Chin
ese).
[7]
Jinq
uan Z
,
L
i
n
a
N, Ch
an
gju
n
J.
A Heur
istic Sche
dul
in
g
Strategy for In
dep
en
dent Ta
sks on Grid
.
Procee
din
g
s
of the E
i
g
h
th Int
e
rnati
ona
l
Con
f
erenc
e
o
n
H
i
g
h
-Performa
nce
Comp
utin
g i
n
Asia-Pac
ifi
c
Regi
on (HP
C
A
S
IA ’05). 200
5.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5559 – 55
66
5566
[8]
Xi
n Lu, Z
i
l
ong
Gu. A load-a
d
a
pative cl
ou
d re
source sch
ed
ul
ing mo
de
l bas
ed o
n
ant col
o
n
y
alg
o
rithm
.
IEEE.
2011.
[9]
Kun L
i
, Gaoc
h
ao
Xu, Gu
an
g
y
u Z
hao, Y
u
sh
u
ang
Do
ng, W
a
ng, D., "Cl
o
u
d
T
a
sk Schedul
i
ng Bas
ed
on
Loa
d Bal
anci
n
g Ant Colo
n
y
O
p
timizati
on.
IEEE.
2011.
[10]
Z
hang Bo, Gao
Ji, Ai Jieqin
g
. Clou
d
Lo
ad
ing
Bala
nce al
gor
ithm.
IEEE.
2011.
[11]
Xi
ao-S
han
He,
Xian-
He S
un,
QoS Guid
ed
Min-Min
He
uri
s
tic for Grid T
a
sk Sche
du
lin
g
. Jo
u
r
an
al
of
Co
mp
uter Scie
nce & T
e
chn
o
l
ogy
. 200
3; (5): 442-
451.
[12]
Sameer
Sin
g
h
Ch
auh
an,
R
C
Jos
h
i. QoS
Guid
ed
He
uri
s
tic Alg
o
rithms
for Grid T
a
s
k
Sche
dul
in
g.
Internatio
na
l Journ
a
l of
Co
mputer App
lic
ations
. 201
0; 2(9)
.
[13]
Meng
Xu, Liz
h
en C
u
i, Ha
i
y
a
n
g
W
ang, Y
anb
i
ng Bi.
A M
u
ltip
l
e
QoS Co
nstra
i
ne
d Sch
edu
lin
g Strateg
y
of
Multipl
e
Workflo
w
s for Clo
ud
Comp
uting.
IEEE
. 2009.
[14]
Frincu ME, Craciu
n C.
Multi-ob
jectiv
e
Meta-h
e
u
risti
cs for Sched
ulin
g Ap
plic
ati
ons
w
i
t
h
Hi
g
h
Avail
abi
lit
y
Re
quir
e
ments an
d Cost C
onstra
i
nts in Multi-C
l
oud Env
i
ro
nme
n
ts
. IEEE.
2012.
[15]
Hon
g
Su
n, Shi-
pin
g
Ch
en, C
h
en Ji
n, Kai Gu
o.
Rese
arch
an
d Simu
latio
n
of
T
a
sk Schedul
i
ng Al
gorith
m
in C
l
ou
d C
o
mp
uting.
T
E
LKO
M
NIKA Indo
ne
sian J
our
nal
of
Electrica
l
En
gi
neer
ing.
20
13; 20(1
1
):
66
64-
6
6
72
[16]
AT
ANASSOV
KT. Intuitionisti
c
fuzz
y
sets
. Fuzz
y
S
e
ts and
S
y
stems. 19
86
; 20(1): 87-96.
[17]
Lazim A
bdu
lla
h, Hui M
e
i L
i
n
g
.Intervalva
lu
e
d
Intu
itio
nistic F
u
zz
y
W
e
i
ghte
d
Entrop
y in Evalu
a
tion o
f
Service Qu
alit
y.
Intern
ation
a
l
Journ
a
l of Inf
o
rmatics a
nd
Co
mmun
icati
o
n T
e
chn
o
lo
gy
(IJ-ICT
)
. 2013;
2(1): 17-2
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.