Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
9
,
No.
3
,
June
201
9
, pp.
1637
~
1644
IS
S
N:
20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v
9
i
3
.
pp
1637
-
16
44
1637
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Demand
-
driven
Gaussia
n
wind
ow
optimi
zation fo
r exe
cuting
preferr
ed
popul
atio
n of j
ob
s in cloud
clusters
Va
ide
hi M
1
,
T
.R
.
Gopal
ak
ri
s
hnan
2
1
Depa
rt
m
ent
of
I
nform
at
ion
Sci
e
nce
and Engi
ne
e
ring,
Da
y
a
na
nd
a
saga
r
Col
le
ge
of
Engi
ne
eri
ng,
Ind
ia
2
Raj
ar
aj
eshwari
Coll
ege of
Enginee
ring
,
Indi
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
un
6
, 201
8
Re
vised
Dec
15
, 2
01
8
Accepte
d
Dec
29
, 201
8
Schedul
ing
is
one
of
the
essenti
al
en
abl
ing
t
echnique
for
Cloud
computing
which
fa
ci
l
itate
s
eff
i
cient
r
esourc
e
u
ti
l
iz
a
ti
on
a
m
ong
the
jobs
sche
dul
ed
fo
r
proc
essing.
How
eve
r,
it
exp
er
ie
nc
es
per
form
anc
e
over
he
ads
due
to
the
ina
ppropri
at
e
pr
ovisioni
ng
of
resourc
es
to
re
que
sting
jobs.
It
is
ver
y
m
uch
essenti
a
l
tha
t
th
e
per
form
anc
e
o
f
Cloud
is
ac
complished
through
int
el
l
ige
n
t
sche
duli
ng
and
al
locati
on
of
resourc
es.
In
t
his
pape
r,
we
propose
th
e
appl
i
ca
t
ion
of
G
aussian
window
where
jobs
of
het
ero
g
ene
ous
i
n
nat
ur
e
a
r
e
sche
du
le
d
in
the round
-
robin
fash
io
n
on
d
iffe
r
ent
Cloud
c
luste
rs.
The
cl
usters
are
heteroge
n
eo
us
in
nat
ure
hav
ing
dat
a
ce
nt
ers
with
var
y
ing
se
ver
ca
p
ac
i
t
y
.
Perform
anc
e
ev
al
ua
ti
on
resul
ts
show
tha
t
th
e
proposed
a
lg
orit
hm
has
enha
nc
ed
the
Q
oS
of
the
comp
uti
ng
m
odel
.
Al
loc
a
ti
on
of
Jobs
to
spe
ci
fi
c
Cluste
rs ha
s
improved
th
e
s
y
stem
throughput a
nd
has
red
uc
ed the l
at
en
c
y
.
Ke
yw
or
d
s
:
Cl
oud
c
om
pu
ti
ng
Dem
and
-
dr
ive
n
Eff
ic
ie
ncy
In
sta
ntane
ous
util
iz
at
ion
Jo
bs
Re
so
urce
util
iz
at
ion
Sche
du
li
ng
Copyright
©
201
9
Instit
ute of
Ad
v
ance
d
Engi
ne
eri
ng
and
Sc
ie
n
ce
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Vaide
hi M,
Dep
a
rtem
ent o
f
I
nfo
rm
ation
Scie
nce a
nd
E
ng
i
neer
i
ng,
Dayana
nd
a
sag
ar Co
ll
ege
of
E
ng
i
neer
i
ng,
Ku
m
arasw
am
y
la
yout,
Ba
ngal
or
e
,
I
ndia
.
Em
a
il
:
vaideh
im
-
ise
@d
ay
anan
da
saga
r.
e
du
1.
INTROD
U
CTION
Gen
e
rati
on
of
a
huge
volu
m
e
of
data
a
nd
re
qu
irem
en
t
of
high
-
s
pee
d
com
pu
ta
ti
on
with
le
ss
inv
est
m
ent
has
pav
e
d
way
to
fast
de
velo
pm
e
nt
of
the
Cl
ou
d
te
ch
no
l
og
y.
Ther
e
is
a
c
omplet
e
trans
form
at
io
n
of
com
pu
ti
ng
wh
e
n
c
om
par
e
d
to
the
tra
diti
on
al
m
et
ho
d.
This
su
cc
e
ssf
ul
m
od
el
is
buil
t
us
in
g
t
he
G
rid
te
chnolo
gy,
V
irtual
iz
at
ion
a
nd
Distrib
uted
com
pu
ti
ng
.
The
Cl
oud
prov
i
des
hi
gh
-
s
peed
proces
sors
f
or
com
pu
ta
ti
on
a
nd
sto
ra
ge
as
a
serv
ic
e.
T
hough
this
m
od
el
is
hig
hly
us
e
d
f
or
com
pu
ta
ti
on
and
sto
rag
e
th
ere
are
sever
al
iss
ues
to
be
ad
dr
esse
d
as
Infr
ast
r
uc
ture
-
as
-
se
rv
ic
e.
Du
e
to
non
-
unif
or
m
and
tim
e
-
var
yi
ng
w
orklo
a
d,
the
resour
c
es
require
d
to
sust
ai
n
the
wor
klo
a
d
is
al
so
var
ia
ble
[1
]
.
Am
azon
,
G
oogle,
et
c.,
inv
e
ste
d
a
consi
der
a
ble
a
m
ou
nt
of
m
on
ey
in
their
data
centers
as
the
y
hav
e
to
m
ai
ntain
the
se
rv
e
r
sto
s
us
ta
in
thei
r
pe
a
k
work
l
oa
d.
T
he
ave
rag
e
util
iz
at
ion
of
ser
vers
was
only
10
%
[
1].
T
hey
then
reali
zed
that
m
erg
ing
di
ff
e
rent
work
l
oa
ds
with
the
c
om
pli
mentary
us
ag
e
pa
tt
ern
s
will
en
han
ce
the
ser
ve
r
ef
fici
ency
a
nd
it
w
ou
l
d
be
a
c
os
t
-
eff
ect
ive
eco
nom
ic
m
od
el
to
ren
t
the
resou
rces
to
the
public
[2
]
.
Am
az
on
la
unc
hed
A
WS
(
Am
azon
We
b
Ser
vices)
util
it
y
com
pu
ti
ng
,
and
a
fter
the
la
un
c
h,
se
ver
a
l
IT
industries
op
te
d
f
or
Cl
oud
c
om
pu
ti
ng
tha
n
inv
est
in
g
on
c
os
tl
y
serv
e
rs
[
2].
As
the
de
m
and
g
re
w,
t
he
c
om
pu
ti
ng
m
od
el
sta
rted
enc
ounteri
ng
seve
re
chall
enges li
ke
job sc
hedulin
g an
d resou
rce a
ll
ocati
on
[3
]
e
xc
lusively
to
c
om
pu
te
r
eal
d
at
a.
The
ser
vice
pr
ov
i
der
has
to
cat
er
to
heter
ogene
ous
jo
bs
and
not
just
a
cl
us
te
r
of
cl
ie
nts
w
hose
requests
are
h
om
og
eneous
in
natu
re.
Chall
e
nges
are
relat
ed
to
flexi
bili
ty
in
IaaS.
T
his
paper
prese
nts
a
novel
appr
oach
usi
ng
the
Ga
us
sia
n
wind
ow
to
op
ti
m
iz
e
the
e
xecu
ti
on
of
pr
efer
red
po
pula
ti
on
of
j
obs
in
cl
ou
d
cl
us
te
rs.
Her
e
the
com
pu
ti
ng
m
od
el
com
pr
ise
s
of
cl
us
te
rs
of
f
our
di
ff
e
r
ent
capaci
ti
es.
Figure
1
pres
ents
a
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
3
,
June
201
9
:
1637
-
1644
1638
con
ce
ptu
al
m
od
el
f
or
a
pp
li
ca
ti
on
of
Ga
us
si
an
wind
ow
.
T
he
m
od
el
incl
udes
of
tw
o
e
ntit
ie
s,
a
J
ob
dis
pa
tc
her
and
Cl
us
te
rs
.
The
m
od
el
es
tim
a
te
s
the
re
qu
i
red
am
ount
of
res
ources
for
c
om
pu
ta
tio
n
of
j
obs
t
o
av
oid
inap
pro
pr
ia
te
use
of the
av
ai
la
ble r
es
ources
.
Figure
1
.
Co
nc
eptual
m
od
el
of cluste
rs wit
h a cl
us
te
r
m
ana
gem
ent syst
e
m
The
pap
e
r
is
struct
ur
e
d
as
f
ollow
s
,
Sect
io
n
I
pr
ese
nts
the
e
voluti
on
of
the
c
om
pu
ti
ng
m
odel
,
feature
s
and
t
he
chall
e
nge
ad
dresse
d
i
n
this
pa
pe
r.
S
ect
ion
I
I
pr
ov
i
des
a
s
umm
ary
of
the
relat
ed
li
te
ratur
e.
Sect
ion
III
pro
vid
es
the
de
ta
il
s
of
the
con
cept
ual
m
od
e
l
and
the
al
gori
thm
fo
r
al
locat
ing
the
res
ourc
es
to
requesti
ng
job
s
.
Larg
e
am
ou
nt
of
job
s
ar
riving
are
cl
assifi
ed
ba
sed
on
prop
e
rt
ie
s
li
ke
si
ze,
res
ource
ut
il
iz
at
ion
per
i
od
a
nd
com
pu
ti
ng
c
ost
.
Perfor
m
ance
evaluati
on
of
the
pro
po
se
d
al
go
rithm
is
discusse
d
in
S
ect
ion
I
V.
Fi
na
ll
y,
Sect
ion
V
c
onc
lud
es
the
pa
per.
This
sect
io
n
bri
efs
ab
out
the
backg
rou
nd
of
the
pro
pose
d
s
yt
e
m
e
m
ph
asi
zi
ng
on
sig
nific
ant
resea
rc
h
work
ca
rr
ie
d
out
to
en
han
ce
t
he
qual
it
y
of
s
evice
in
the
cl
oud
syst
em
.
Sc
hedulin
g
an
d
Re
so
urce
al
loc
at
ion
in
the
heter
ogen
eous
cl
oud
e
nv
i
ronm
ent
is
an
op
e
n
res
earch
c
halle
nge
w
he
re
m
a
ny
resea
rch
e
r
s
an
d
academ
ic
ia
ns
a
re
w
orkin
g
to
w
ar
ds
to
e
nhanc
e
the
perform
a
nce
of
the
com
pu
ti
ng
m
od
el
.
Fo
rm
the
su
r
ve
y
it
is
ob
s
er
ved
t
hat
the
sch
ed
ulin
g
te
chn
i
qu
e
s
has
an
im
pact
on
com
pu
ta
ti
on
co
st,
resou
rce
uti
li
zat
ion
tim
e,
e
nerg
y
util
iz
at
ion
and
Q
oS,
eff
ic
ie
nt
sche
du
li
ng
al
so
reduces
the
job
re
j
ect
io
n
rat
io.
T.
R.
Gopal
akr
is
hn
a
n
Nair
et
a
l.,
in
their
w
ork
s
ay
that
it
is
es
sentia
l
to
i
den
t
ify
the
tre
nds
of
dif
fere
nt
re
qu
e
st
stream
s
in
e
ver
y
cat
e
go
ry
by
auto
cl
assifi
cat
ion
a
nd
orga
ni
ze
pr
e
-
al
locat
ion
of
res
ource
s
in
a
pr
e
dicti
ve
way
to
reduc
e
the
num
ber
of
jo
bs
bein
g
reject
ed
and
al
so
re
duct
ion
in
c
os
t
per
ta
sk
com
pleti
on
[
4].
M.
Me
z
m
az
e
t
al
.
in
their
wo
r
k
ha
s
inv
est
igate
d
th
e
prob
le
m
of
sche
du
li
ng
the
prob
le
m
pr
ecedence
-
co
ns
tr
ai
ned
par
al
le
l
app
li
cat
io
ns
on
th
e
heter
og
e
ne
ou
s
com
pu
ti
ng
syst
e
m
(Clou
d
c
om
pu
ti
ng),
in
their
w
ork
the
y
hav
e
pro
pos
ed
a
ne
w
pa
ra
ll
el
bi
-
obj
ect
ive
hybr
id
ge
netic
al
go
rithm
that
tak
es
into
acc
ou
nt
the
ta
sk
c
om
ple
ti
on
tim
e
and
al
s
o
m
in
i
m
iz
ed
energy
c
on
s
um
pt
ion
[5
]
.
Mi
ng
s
ong
Che
n
et
al
.,
say
that
due
to
the
e
xi
ste
nce
of
res
ource
va
ria
ti
on
s
,
it
is
a
chall
enge
for
c
loud
w
orkf
l
ow
resour
ce
al
loc
at
ion
strat
egies
to
gu
a
ran
te
e
a
reli
able
Qo
S
.
They
al
so
say
that
it
is
har
d
t
o
pr
edict
their
pe
rfor
m
ance
under
var
ia
ti
ons
becau
se
of
la
ck
of
acc
urat
e
m
od
el
li
ng
and
evaluati
on m
eth
ods
[
6].
Hsu
M
on
K
yi
et
al
.,
say
that
in
cl
ou
d
c
om
pu
ti
ng
syst
em
s
sched
ulin
g
and
al
locat
io
n
of
virtua
l
resou
rces
an
d
virtu
al
m
achine
are
chall
e
ng
e
s.
To
a
ddres
s
this
issue
,
they
hav
e
pro
posed
an
al
gorithm
wh
i
c
h
pro
vid
es
e
ff
ect
ive
an
d
e
ff
ic
ie
nt
res
ource
al
l
ocati
on.
T
hey
hav
e
u
se
d
Stoc
hastic
Ma
rko
v
m
od
el
to
m
eas
ur
e
t
he
scal
abili
ty
and
tract
abili
ty
of
infr
a
struct
ur
e
resou
rce
in
pri
vate
cl
ouds.
Their
c
on
t
rib
ut
ion
ha
s
f
ocus
ed
on
enh
a
ncin
g
the
syst
e
m
per
for
m
ance
by
enabl
ing
the
res
ponse
tim
e
[7
]
W
e
xin
Li
et
al
.,
hav
e
pro
posed
a
j
oin
t
op
ti
m
iz
ation
m
od
el
,
this
c
hoose
s
the
re
qu
est
al
locat
ion
po
li
cy
s
uch
th
at
the
prov
i
der
gain
s
high
ba
ndwidt
h
util
iz
at
ion
at
it
s
datace
nte
rs,
and
each
us
e
r
exp
e
riences
a
low
delay
[
8].
Pandaba
et
al
.,
in
their
w
ork
say
that
cl
oud
in
frast
r
uc
ture
c
om
pr
ise
s
of
se
ver
al
da
ta
centers,
an
d
the
c
us
tom
er
need
a
sli
ce
of
the
com
pu
ta
t
ion
al
powe
r
over
a
scal
able
netw
ork
.
They
say
de
li
ver
y
of
re
sources
a
re
done
in
an
el
ast
ic
way.
The
c
halle
nge
inv
est
igate
d
by
the
m
is
the
wait
tim
e
exp
erie
nced
by
the
cust
om
ers.
The
resea
rc
he
rs
ha
ve
pro
pose
d
a
m
od
ifie
d
Ro
und
Ro
bin
al
gorit
hm
that
reduce
s
the w
ai
t
tim
e,
there
b
y
im
pr
ovin
g
the
pe
rform
ance
[
9]
Mu
bar
a
k
et
al
,
hav
e
m
ade
a
stud
y
on
ta
sk
sche
du
li
ng
al
gorithm
s.
Th
e
resea
rc
her
s
hav
e
e
nhanced
the
Min
-
Mi
n
al
gorithm
to
enh
a
nce
t
he
ta
s
k
c
om
pleti
on
pe
rio
d.
T
he
a
uthors
say
that,
th
rou
gh
t
he
e
xp
e
rim
ental
analysis,
the
pro
po
se
d
al
go
rithm
has
pr
oduce
d
a
bette
r
Ma
ke
sp
an
a
nd
im
pr
oved
resou
rces
util
iz
at
ion
[
10
]
.
S
te
fano
Ma
rron
e
et
.al,
hav
e
pro
posed
a
m
od
el
-
d
rive
n
ap
proac
h
f
or
the
autom
at
ic
negotia
ti
on
a
nd
res
ource
al
lo
cat
io
n
for
avail
abil
it
y
of
crit
ic
al
cl
oud
ser
vices.
T
he
authors
hav
e
use
d
Ba
ye
sia
n
ne
twork
to
e
valuate
the
avail
abili
ty
of r
es
ources
fo
r
crit
ic
al
servic
es [11
]
.
Th
ough
m
any
researc
h
has
bee
n
car
ried
ou
t
t
o
en
ha
nc
e
the
res
ourc
e
util
iz
at
ion
in
the
cl
ou
d
env
i
ronm
ent, ther
e
f
e
w
a
reas
to b
e
foc
us
ed
to
inc
rease t
he qu
al
it
y o
f
se
rv
i
ce in clo
ud e
nv
iro
nm
ent.
The
c
ur
ee
nt sy
stem
s less fo
c
us o
n
a
ppr
oach
e
s to
e
nhance
1.
Lat
ency o
f
t
he c
om
pu
ti
ng
m
odel
2.
Thro
ugh
put
Im
pr
ov
i
ng thes
e two pa
ram
et
e
rs would
enha
nc
e b
et
te
r uti
li
zat
ion
of r
es
our
ces in cl
oud
e
nvir
on
m
ent
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
Dema
nd
-
dr
iv
e
n Gaussi
an
wi
ndow
opti
miza
t
ion
f
or
exec
uting pr
ef
erred
popu
lati
on
of j
obs
…(
Vaid
e
hi
M
)
1639
This
w
ork
fo
c
us
es
on
en
hanci
ng
the
syst
em
per
form
ance
by
op
ti
m
a
l
util
iz
at
ion
of
the
avail
abl
e
resou
rces.
T
o
achieve
this
,
we
pro
pose
a
con
ce
ptu
al
m
od
el
as
sho
wn
i
n
F
ig
ur
e
1.
Th
e
m
od
el
com
pr
ise
s
of
cl
us
te
rs
w
hich
is
a
set
of
m
achines
pack
e
d
into
rac
ks
.
These
cl
us
te
r
s
are
connecte
d
with
high
ba
ndwidt
h
cl
us
te
r
netw
or
k.
Cl
us
te
rs
a
re
m
anag
ed
by
the
Cl
us
te
r
m
a
nag
em
ent
syste
m
wh
ic
h
al
locat
es
j
obs
to
m
achines
.
Jo
bs
gen
e
rall
y
ha
ve
a
set
of
resou
rce
r
eq
uir
e
m
ent
for
sc
he
du
li
ng
or
packi
ng
t
he
ta
s
ks
i
n
the
m
achine
ei
ther
for
sto
ra
ge
or
execu
ti
on.
T
he
Ga
us
sia
n
wi
ndow
m
od
el
is
app
li
ed
to
t
he
pro
po
se
d
m
odel
to
accom
plish
best
util
iz
at
ion
of a
vaila
ble r
e
sour
ces. T
he Gau
ss
ia
n
wi
ndow is
app
li
ed
to dif
fe
ren
t C
lu
ste
rs.
The
m
achines
are cluste
red d
ependin
g o
n diff
e
ren
t
data
ha
nd
li
ng a
nd
pro
cessi
ng s
peed.
Limi
tati
on
s in
the exist
ing
Resource
Al
locat
ion Model
Ma
ny
of
the
r
esearche
rs
w
orkin
g
on
sc
heduling
an
d
res
ource
al
locat
ion
ha
v
e
pro
po
se
d
new
al
gorith
m
s
an
d
com
pu
ti
ng
m
od
el
but
thei
r
w
ork
does
no
t
em
ph
asi
s
m
uch
on
cl
us
te
rin
g
t
he
m
achines
acc
ordin
g
t
o
com
pu
ta
ti
on
or stor
a
ge re
qu
ir
e
m
ents.
App
li
catio
n
and w
or
kin
g o
f
t
he a
l
gorit
hm
The
propose
d
al
gorithm
is
a
pp
li
ed
t
o
a
set
of
sam
ple
d
at
a
show
n
Ta
ble
1.
T
he
ta
ble
com
pr
ise
s,
of
Jo
bI
D,
Ti
m
e
stam
p
assigne
d,
siz
e
of
the
job
a
nd
Ma
chine
I
D
as
sign
e
d
t
o
eac
h
com
pu
ti
ng
m
achine
an
d
fin
al
ly
the
cl
us
te
rs form
e
d wit
h
a
n Id as
sign
e
d.
Table
1.
Param
et
ers
R
el
at
ed
t
o
J
obs S
c
he
du
l
ed
Sl.no
.
Jo
b
I
D
Ti
m
e
Sta
m
p
Jo
b
Size
Machin
e I
d
(Static
I
P)
Clu
ster I
d
1
J
1
T1
6
0
0
KB
2
0
0
.168.2
.2
Clu
ster0
2
2
J
2
T2
5
0
0
MB
1
9
9
.170.5
.1
Clu
ster0
1
3
J
3
T3
6
0
0
KB
2
0
0
.168.2
.3
Clu
ster0
2
4
J
4
T4
2
0
0
KB
2
0
0
.168.2
.4
Clu
ster0
2
5
J
5
T5
7
0
0
MB
1
9
9
.170.5
.3
Clu
ster0
1
6
J
6
T6
2
9
9
KB
2
0
0
.168.2
.5
Clu
ster0
1
The
Cl
us
te
r
m
anag
em
ent
syst
e
m
(CMS)
classifies
the
j
obs
base
d
on
th
e
resour
ce
re
quirem
ent
fo
r
processi
ng
an
d
then
distri
bu
te
s
to
a
ppr
opriat
e
cl
us
te
rs
.
T
he
Roun
d
-
robi
n
s
cheduli
ng
is
use
d
for
the
exec
utio
n
of the
job
s
in
c
lusters.
In
it
ia
ll
y,
the
j
obs
are
cl
assi
fied
by
CM
S
as
a
fre
e
pr
iorit
y,
pr
od
uctio
n
pri
or
it
yand
moni
tor
pr
i
or
it
y
job
s
.
The
f
ree
pr
i
or
it
y
jo
bs
us
e
m
ini
m
um
reso
ur
ces
for
c
om
pu
ta
tio
n,
a
nd
the
com
pu
ta
ti
on
al
cost
is
com
par
at
ively
low,
t
he
produ
ct
ion
pr
io
rity
job
s
ha
ve
th
e
hi
gh
e
st
pr
i
or
it
y,
the
CM
S
sees
t
o
that
the
se
jo
bs
are
no
t
den
ie
d
of
t
he
re
quest
ed
r
eso
ur
ces
,
an
d
t
hey
are
al
s
o
not
al
locat
ed
to
ov
e
rloa
de
d
m
a
chines
.
T
his
ensure
s
that
load
bala
ncin
g
is
ta
ke
n
care
of
the
pro
posed
m
od
el
.
The
fr
ee
pr
i
or
it
y
job
s
a
re
ta
ken
ca
re
of
by
the
m
on
it
or
p
ri
or
it
y jobs
to e
ns
ur
e resour
c
es.Ea
ch
jo
b
has
a ti
m
e sta
m
p
(t
s
),
job
Id
(J
i
)
a
nd a
co
m
par
ison
operat
or
.
The
c
om
par
iso
n op
e
rato
r
is
great
er th
a
n or l
ess tha
n.
Resources
and U
nits
CPU
-
num
ber
of c
or
es/
sec
ond
Mem
or
y
-
byte
s
Disk
Sp
a
ce
-
byte
s
Disk
ti
m
e fr
act
ion
(I
/
O
in
sec
onds
/ sec
onds)
[12
]
2.
ALGO
RITH
M
I
MPLEME
NTATIO
N
a.
Assum
ption
: w
it
h
ref
e
re
nce to t
he pr
opos
e
d m
od
el
in
F
ig
ure 1, the cl
us
te
r
s 1
t
o 4 ar
e
C1
, C
2,
C
3
a
nd C4
Assum
ption
s:
1.
C1: data
ha
nd
li
ng capa
bili
ty
in
Peta
byte
s
C2: data
ha
nd
li
ng cap
a
bili
ty
in
Te
rab
yt
es
C3: data
ha
nd
li
ng capa
bili
ty
in Giga
byte
s
C4: data
ha
nd
li
ng capa
bili
ty
in
Me
gab
yt
es
2.
Jo
b I
d=
1,2,3, a
ssign
e
d base
d on their
ar
rival
r
at
e as
J
1
, J
2
,
J
3
…...
J
n
3.
Synchr
on
iz
at
io
n
am
on
g t
he
Cl
us
te
rs
4.
Cl
us
te
r
fail
ur
e
ta
ken
ca
re a
utom
at
ic
ally
Step 1
:
J
ob
1
if
reso
ur
ce re
quire
me
nt in
Pet
abyt
es size
then
Step
2: Check f
or
res
ources
av
ailab
le
at t
ha
t
insta
nce
of
ar
r
iv
al
If a
v
aila
ble sc
hedule
Else
Step 3
:
Check
f
or
res
ources i
n C
2 a
nd C3
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
3
,
June
201
9
:
1637
-
1644
1640
If a
v
aila
ble
Step 4
: Sp
li
t t
he
job in ter
abyt
es size
and gig
ab
y
te
s
size
Syn
c
C2
and
C
3
Sch
e
dule
Step 5
:
Else
bl
ock th
e
Job
1
Step 6
:
Check
wait
ti
me
of J
ob
1
and
c
heck
reso
ur
ce
av
ailab
le
If a
v
aila
ble
Step 7
: Unbloc
k Jo
b
1
and
sc
he
du
le
Re
peat f
or all
job
s
b.
Applic
at
ion o
f Gaus
sia
n w
in
dow
In
t
his
pa
per,
we
im
ple
m
ent
the
Ga
us
sia
n
window
to
a
set
of
job
s
.
The
un
i
qu
e
pr
op
e
rty
of
the
al
gorithm
is
th
at
j
us
t
the
prev
io
us
histo
ry
of
the
job
siz
e
[
8]
is
su
ff
ic
ie
nt
t
o
pre
dict
the
r
eso
ur
ce
requir
e
m
ent
for
the
cu
rr
e
nt
j
ob.
K
nowing
the
Me
an
“
µ”
the
aver
a
ge
require
d
res
ources,
a
nd
t
he
sta
nd
ar
d
de
viati
on
σ
2
wh
ic
h
in
dica
te
s
the
act
ual
r
eso
ur
ce
util
iz
ed
by
t
he
jo
b
ea
rlie
r
will
ena
ble
the
CM
S
to
al
locat
e
resource
for
the
cu
rr
e
nt
j
ob
in
e
xec
utio
n
with
ou
t
wasti
ng
tim
e
fo
r
c
om
pu
ta
ti
on
of
res
ource
re
qu
irem
ent.
Hen
c
e
this
appr
oach al
so
will
incr
ease
th
e thro
ughput
of the
syst
em
.
In our resea
rc
h, we
u
s
e the
infor
m
at
ion
of t
he
r
es
ource size
al
locat
ed
f
or e
xecu
ti
on
of the
previ
ou
s
jo
b.
c.
Gau
s
sia
n win
dow
for p
re
dicti
on
Gau
s
sia
n
distr
ibu
ti
on
is
a
po
we
rful
too
l
app
li
ed
e
xp
a
nsi
vely
to
pr
ob
lem
s
li
ke
reg
r
ession
a
nd
cl
assifi
cat
ion
.
In
the
Gaussi
an
proces
s
a
pri
or
pro
ba
bili
ty
distrib
ution
is
assum
ed
init
i
al
ly
ov
er
the
va
rio
us
functi
ons
po
s
s
ible
to
descr
i
be
the
process
of
ge
ner
at
in
g
data,
a
nd
a
poste
rio
r
pro
ba
bili
ty
distribu
t
ion
is
ob
ta
ine
d
after
gaining
knowl
edg
e
ab
out
the
obser
ve
d
valu
es.
T
he
poste
ri
or
im
pr
ov
es
th
e
knowle
dge
of
the
ob
s
er
ved ove
r t
he
pri
or.
[
]
=
ⅇ
-
y
Wh
e
re y=
[
−
]
2
2
2
W
it
h resp
ect
to
G
a
us
sia
n wi
ndow
The param
et
ers
a= de
fines
t
he req
uirem
ent o
f
the
resou
rces
x=
T
he
act
ual
al
locat
ed
re
sou
rce
c
2
=def
ine
s the
real res
ources
util
iz
ed
b=
the
av
e
ra
ge
r
es
ource
requi
red
d.
Si
m
ulati
on
set
up and
predict
ion o
f res
ults
To
im
ple
m
ent
the
pro
posed
a
lgorit
hm
to
ac
hieve
the
res
ults,
heter
og
e
ne
ou
s
Cl
us
te
rs
al
ong
with
a
com
m
un
ic
at
ion
netw
ork
us
ing
a
TC
P
pr
oto
c
ol
was
buil
t.
The
Cl
ust
ers
we
re
c
re
at
ed
us
i
ng
V
Mware.
W
i
res
hark t
oo
l
w
as
us
e
d
t
o
m
on
it
or t
he Dat
a transm
issi
on
a
m
on
g
to t
he
Cl
us
te
rs
.
3.
RESU
LT
A
N
ALYSIS
Thr
ee
d
i
ff
e
ren
t
cases
wer
e
conside
red to
pro
ve
the
appli
cat
ion o
f
t
he pr
opose
d
al
gorithm
, h
ere
a.
Ca
se I
: B
est
Util
iz
at
ion
o
f
av
a
il
able resou
rce
s b
y
requeste
d Jobs
Figure
2
dep
ic
t
s
the
best
util
iz
at
ion
of
the
al
locat
ed
re
sourc
es;
her
e
the p
r
opos
e
d
al
gorit
hm
cl
assifi
es
the
jo
bs
a
s
per
the
res
ources
requested
.
It
c
an
be
obser
ve
d
that
the
area
unde
r
t
he
c
urve
sh
ows
the
util
iz
at
ion
of the all
ocate
d resou
rces.
The
ap
plic
at
io
n
of
the
Ga
us
s
ia
n
wind
ow
ha
s
al
so
ena
bled
to
en
han
ce
the
la
te
ncy
and
th
rou
ghput
of
the
syst
em
.
The
Fig
ur
e
3
a
nd
Fig
ure
5
il
lustrate
s
the
la
te
ncy
an
d
Fi
gur
e
4
il
lustrate
s
the
thr
ough
pu
t
.
Th
e
la
te
ncy
per
io
d.
W
it
h
re
fer
e
nc
e
to
Fig
ur
e
3
tim
est
a
m
p
“0n
s”
dep
ic
ts
th
e
st
art
of
e
xec
utio
n
of
a
ta
sk,
the
ta
sk
wait
s appro
xim
at
ely f
or “
6ns” to l
oad on t
o t
he
m
achine a
nd co
m
plete
s the e
xecu
ti
on in
556ns.
Be
fore
sche
du
l
ing
,
cl
ust
erin
g
of
the j
obs b
as
ed
on
their
jo
b
siz
e
based
o
n
t
he
pro
posed
al
gorithm
th
e
la
te
ncy
of
the
syst
e
m
is
co
m
par
at
ively
im
p
rove
d.
Fi
gure
6
re
pr
e
sents
t
he
thr
oughput
of
the
propose
d
syst
e
m
.
It
is
obser
ved
that
sche
duli
ng
t
he
j
obs
to
a
ppr
opriat
e
cl
ust
ers
has
reduce
d
t
he
j
ob
re
j
ec
ti
on
s
a
nd
sta
rvat
ion
.
The
sc
he
du
le
d jobs
h
a
ve fully
u
ti
li
zed th
e
all
oc
at
ed res
ourc
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
Dema
nd
-
dr
iv
e
n Gaussi
an
wi
ndow
opti
miza
t
ion
f
or
exec
uting pr
ef
erred
popu
lati
on
of j
obs
…(
Vaid
e
hi
M
)
1641
Figure
2.
O
ptim
al
reso
urce
ut
il
iz
ation
Figure
3.
Lat
en
cy
o
f
the
pro
posed
syst
em
Figure
4
.
Re
source
util
iz
at
ion by the
all
ocate
d jobs
in
a
ppr
opriat
e cluste
rs
Figure
5
.
Ro
undtrip f
ro
m
the
pro
po
se
d
syst
e
m
Figure
6
.
Th
r
ough
pu
t
of the
s
yst
e
m
w
hen jo
b
Size
is
prop
ort
ion
at
e t
o
the
re
qu
e
ste
d res
ource
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
3
,
June
201
9
:
1637
-
1644
1642
b.
Ca
se I
I:
U
nd
e
r
util
iz
at
ion
of R
eso
ur
ces
whe
n job si
ze is l
ess
than
t
he reso
ur
ce al
locat
ed
Figure
7
s
how
s
the
util
iz
at
ion
of
res
ources
by
j
obs
al
loc
at
ed
to
m
achi
nes.
In
this
c
a
se
the
act
ua
l
resou
rce
re
quir
ed
f
or
com
pu
t
at
ion
is
not
est
i
m
at
ed
pr
io
r
to
sche
du
li
ng.
T
he
area
unde
r
t
he
cu
r
ve
de
pic
ts
the
util
iz
at
ion
of
a
ll
ocated
res
our
ces.
Th
ough
th
e
com
pu
ta
ti
on
is
com
plete
,
the
resou
rces
are
no
t
fu
ll
y
util
iz
ed
.
The
th
rou
ghpu
t
of
the
syst
em
is
red
uc
ed
by
15
-
20
%
.
Fig
ure
8
sho
ws
the
aver
a
ge
re
duce
in
thr
oughput
of
the
syst
e
m
.
W
e
ca
n
obse
rv
e
th
at
tho
ug
h
the
com
pu
ta
ti
on
is
com
pleted
the
al
locat
ed
res
ources
are
not
com
plete
ly
u
ti
l
iz
ed.
Figure
7.
U
nde
ru
ti
li
zat
ion
of
al
locat
ed
re
sou
rces
Figure
8.
Th
r
ough
pu
t
of the
s
yst
e
m
w
hen
jo
b
siz
e
is l
ess
th
an
the
all
ocate
d resou
rce size
c.
Ca
se
II
I:
In
c
om
ple
ti
on
of
ex
ecuti
on
du
e
to
def
ic
it
of
re
qu
i
red
res
ources:
The
m
achines
are
no
t
cl
us
te
r
ed
and res
ources
are all
ocated
wi
tho
ut e
stim
at
i
ng the
re
qu
ire
d res
ources
f
or
c
om
pu
ta
ti
on
In
this
case,
Fi
gure
9
r
ep
rese
nts
res
ource
ut
il
iz
at
ion
wh
ere
the
j
ob
siz
e
is
gr
eat
er
tha
n
the
res
ource
s
al
locat
ed.
Here
the
re
source
s
are
al
locat
e
d
to
a
tra
diti
on
al
c
om
pu
ti
n
g
m
od
el
w
here
the
re
source
s
ar
e
no
t cl
us
te
re
d.
Figure
9.
De
fici
t of
res
ources
to
al
locat
ed
jo
bs
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
Dema
nd
-
dr
iv
e
n Gaussi
an
wi
ndow
opti
miza
t
ion
f
or
exec
uting pr
ef
erred
popu
lati
on
of j
obs
…(
Vaid
e
hi
M
)
1643
4.
CONCL
US
I
O
N
Eco
no
m
ic
data
com
pu
ti
ng
an
d
stora
ge
has
pav
e
d
way
f
or
IT
ind
ust
ries
and
re
searc
hers
to
i
m
pr
ove
the
existi
ng
sy
stem
by
ref
ram
ing
an
d
re
desi
gn
i
ng
the
e
xisti
ng
te
ch
nolo
gy.
In
our
resea
rc
h
w
ork
we
pro
po
s
ed
a
Con
ce
ptu
al
m
od
el
of
Fu
tu
re
Cl
oud
Cl
ust
er.Th
e
a
pp
li
c
a
ti
on
of
the
G
aussian
window
to
this
m
od
el
has
enh
a
nce
d
t
he
s
yst
e
m
per
f
or
m
ance.
From
the
res
ults
it
can
be
obse
rv
e
d
t
hat
the
al
locat
ed
resou
rces
a
re
best
util
iz
ed,
the
pro
po
se
d
al
go
rithm
a
lso
preven
ts
resou
r
ce
con
te
ntion.
The
sim
ulati
o
n
res
ults
sho
w
the
i
m
pr
ovem
ent
i
n
thr
ough
pu
t
a
nd
reduce
in
la
te
ncy.
Furthe
r
the
propose
d
syst
e
m
can
su
pport
op
ti
m
al
energy
util
iz
at
ion
whi
ch wil
l be ca
rr
i
ed
as
an exte
nsi
on
.
REFERE
NCE
S
[1]
Saij
ie
Hu
ang,
Mingsong
Chen,
Xiao
L
iu,
De
hui
Du,
Xiaoho
ng
Chen,
"V
ariati
on
-
Aw
are
R
e
source
Allocation
Eva
lu
at
ion
for
C
loud
W
orkflows
Us
ing
Stat
isti
cal
Model
Ch
ec
ki
ng
,
"
2014
IE
EE
Inte
rnational
Co
nfe
renc
e
on
Bi
g
Data
and
Cloud
Computing
(
BdCl
oud)
,
vol. 01, n
o
.
pp
.
201
-
208
,
2014,
doi
:10.
110
9/BDCloud,
48
,
2014
.
[2]
T.
R
.
Gopal
akr
ishnan
Nair
,
M.
Vaid
ehi
,
“
Eff
i
ci
en
t
Resourc
e
Arbitra
t
ion
and
Alloc
a
ti
on
Str
at
eg
ie
s
in
C
lou
d
Com
puti
ng
thro
ugh
Virtua
l
iz
a
tion
,
”
2011
In
te
rn
ati
onal
Conf
ere
nce
on
Cloud
C
omputing
and
In
te
lligen
ce
S
ystem
s
(
CCIS
)
,
China
,
I
EE
E
Xplore
201
1
.
[3]
T.
R.
Gop
al
akr
ishnan
Nair
,
P
.
Ja
y
are
kha
,
"
Pre
-
allocation
Str
at
eg
ie
s
of
Com
puta
ti
on
al
R
esourc
es
in
Clou
d
Com
puti
ng
using
ada
pti
v
e
Res
onanc
e
Th
eor
y
-
2
,
"
Inte
rnationa
l
Journal
on
Cloud
Computing:
Serv
ices
and
Archi
t
ec
tu
re(
IJC
CSA)
,
Vol.
1,
No
.
2,
Augus
t
2011
.
[4]
M.Mez
m
az
,
N.
Mela
b,
Y.Kessa
ci
,
Y.C.Lee,
E.G.
Talbi,
A.Y.
Z
om
a
y
a
,
D.
Tu
y
tte
ns
,
“
A
Para
ll
e
l
Bi
-
objecti
v
e
H
ybrid
Meta
heur
isti
c
fo
r
Ene
rg
y
-
awa
r
e
Schedul
ing
for
Cloud
Com
put
ing
S
y
stems
,
"
El
sev
i
er,
Journal
of
Parall
el
an
d
Distrib
ute
d
Com
puti
ng
71(
2011)
,
pp
-
1497
-
1508
,
2011
.
[5]
Mingsong
Chen,
Saij
ie
Huang
,
Xin
Fu,
Xiao
Liu,
Jifeng
He
,
“
Stat
ist
ical
Model
Chec
king
-
Bas
e
d
Eva
luation
an
d
Optimiza
ti
o
n
for
Cloud
W
orkflow
Resourc
e
Alloc
ation
,
”
I
EE
E
Tr
ansacti
ons
on
Cloud
Computing
,
Vol.
pp
-
99,
d
oi
:
10.
1109/T
CC
.
20
16.
2586067
,
201
6.
[6]
Hs
u
Mon
Ky
i
and
Thi
nn
Thu
Naing
“
Stocha
stic
Markov
Model
Approac
h
For
Eff
ic
ie
n
t
Virtua
l
Ma
chi
nes
Schedul
ing
On
Privat
e
Cloud
,
”
Inte
rnational
Jo
urnal
on
Cloud
Computing
Serv
ic
es
and
Archi
tecture
(
IJCCSA
)
,
Vol.
1,
No.3, November
2
011
.
[7]
W
enxi
n,
Heng
Qi,
Kegiu
L
i,
Julong
L
an,
“
Joint
opti
m
i
za
t
ion
of
Bandwidt
h
f
or
Provider
and
Delay
for
Us
er
in
Software
Def
ined Dat
a
Cen
te
rs
,
”
IEEE
Tr
ansactions
on
Cloud
Co
mputing
,
Vol
5,
No 2,
Apri
l
-
June
2017
.
[8]
Pandaba
Pradh
a
n,
Praful
la
Ku
.
Behe
ra
,
B
.
N.B.
R
a
y
“
Modifie
d
Round
Robin
Al
gorit
hm
for
Res
ourc
e
Al
loc
a
ti
on
in
Cloud
Com
puti
n
g
,
”
Proc
edia
Co
m
pute
r
Scie
n
ce, IS
SN
:
1877
-
0509,
Vol
:
85
,
PP
: 878
-
890,
2016
.
[9]
Mubara
k
Hal
ad
u,
“
Optimizi
ng
Ta
sk
Schedulin
g
and
Resourc
e
al
lo
ca
t
ion
in
C
loud
Data
Cente
r
using
Enha
n
ced
Min
-
Min Al
gorithm
,
”
IOSR Journal
of
Computer
Engi
ne
ering, (
IOSR
-
JCE
)
,
Vol:
18,
PP
18
-
25,
20
16
.
[10]
Stef
ano
Marron
e,
Robert
o
Nard
one,
“
Autom
at
ic
Resourc
e
Allocati
on
for
High
a
vai
l
abi
lit
y
Clou
d
Servic
es
,
”
2nd
Inte
rnational
W
orkshop on
Computati
onal
An
ti
fr
agil
ity
a
nd
Antif
ragile
Engi
ne
eri
ng
2015
,
2015.
[11]
Hongbing
W
ang,
Zu
li
ng
Kang
,
Li
e
W
ang
,
"P
erf
orm
anc
e
-
Aw
are
Cloud
Resourc
e
Alloc
a
ti
on
Vi
a
Fitne
ss
-
Ena
bl
ed
Aucti
on
,
"
IEEE T
rans
act
ions o
n
Parall
el and
Dis
tribut
ed
Syste
ms
,
Vol
.
27
,
No.4
,
April
2016
.
[12]
Mohit
Dhingra
,
J.L
akshm
i,
S.K
.
Nand
y
,
Chira
n
jib
Bhat
t
ac
har
a
y
a
,
K.Gopina
th
,
“
El
asti
c
Resourc
es
Fram
e
W
ork
for
Iaa
S,
Preserv
ing
Perform
anc
e
S
LA’s
,
”
Proc
ee
d
i
ngs,
6th
Int
erna
ti
onal
Con
fe
ren
ce
on
C
loud
C
omputing
,
Sant
a
Cla
ra
,
C
al
iforn
ia,
US
,
June
2013
.
[13]
Guilhe
rm
e
Da
C
unha
Rodrigue
s,
Rodrigo
N.
Calheiros,
Vini
cus
Ta
var
es
Guim
ara
es,
Gled
erson
Le
ss
a
dos
Santos,
Marc
io
Ba
rbosa
de
Carv
al
ho
,
L
i
sandro
Za
m
benedet
i
Granv
il
l
e,
L
ia
ne
Ma
rga
rid
a
Rocke
nba
ch
T
ar
ouco,
R
aj
kum
ar
Bu
y
y
a
”, S
y
m
posium
on
Applie
d
Com
puti
ng,
AC
M,
Pis
a
It
aly
20
16
.
[14]
Anton
y
Thomas
,
Kris
hnal
a
l
G,
Jaga
th
y
Ra
y
V.P,
“
Credi
t
Based
Scheduling
Algori
thm
in
Cloud
Com
puti
ng
Envi
ronm
ent
,
”
I
nte
rnational
Co
nfe
renc
e
and
Co
mm
unic
ati
on
Technol
ogi
es
(
ICI
CT
2014)
doi:
10.
1016
/j
.
pr
ocs.
2015.
02
.
162
Scie
n
ce Direct
,
2014
.
[15]
Nare
ndra
Kum
ar,
Sw
at
i
Saxen
a
,
“
A
Prefe
r
enc
e
-
base
d
Resour
ce
Allocati
on
in
Cloud
Com
puting
S
y
stems
,
”
3rd
Inte
rnational
Co
nfe
renc
e
on
Re
c
ent
Tr
ends
in
C
omputing
2015
(
ICRTC
-
2015)
,
doi:
10.
1016
/j.pro
cs.
2015
.
07.
375
Scie
nc
e
Dir
ec
t
,
2015.
[16]
Alexa
nder
Ngen
zi
,
Selv
ara
n
i
R
,
Such
it
hra
R
,
“
FDMC:
Fram
ework
for
Dec
isi
o
n
Making
in
C
loud
for
Eff
icie
nt
Resourc
e
Man
a
gement
,
”
Int
ernati
onal
Journal
of
E
le
c
tric
al
an
d
Computer
Eng
ine
ering
(
IJE
CE
)
,
Vol.
7
,
No.
1,
,
pp.
496~504 ISS
N:
2088
-
8708,
DO
I:
10.
11591/
ij
e
ce
.
v7
i1.
pp496
-
50
,
Februa
r
y
201
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
3
,
June
201
9
:
1637
-
1644
1644
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Mr
s.Vaidehi
M
obti
an
ed
B.
E.
(
El
e
ct
roni
cs
and
Com
muni
cation)
and
M.E
.
(Inform
at
ion
Te
chno
log
y
)
de
gre
es
from
Ban
gal
ore
Univer
sit
y
in
th
e
y
e
ar
2
001
and
2007
r
espe
ctively
.
She
is
per
suing
her
Ph.D
in
Visw
eswaray
Technol
ogi
c
al
Univer
sit
y
,
Karna
ta
k
a,
Ind
a
i.
She
is
cur
ren
tly
working
as
As
st.Profe
ss
or
in
De
par
tment
of
Info
rm
at
ion
Sci
ence
and
Engi
n
ee
rin
g,
Da
y
a
nanda
sa
gar
Coll
ege
of
Eng
i
nee
ringB
anga
lor
e.
The
aut
hor
ha
s
serve
d
as
Resea
rch
As
socia
t
e
i
n
Resea
rch
Indu
str
y
and
Inc
ub
at
io
n
Cent
er
,
DS
I.
Her
Resea
r
ch
In
te
r
e
st
s a
re
C
loud
Co
m
puti
ng,
Com
pute
r
Ne
tworks.
dm
.v
ai
deh
i@
gm
ai
l.co
m
,
vaid
ehim
-
ise
@d
ay
anandasa
gar.ed
u
Dr.
T.
R.
Gopalakri
sh
nan
Nai
r
,
obtained
his
M.T
ec
h
degr
e
e
from
India
n
Instit
ute
Scie
n
c
e,
Banga
lor
e
and
his
Doctor
ate
in
Com
pute
r
Science
and
Engi
ne
eri
ng.
He
is
a
m
ember
in
seve
ral
profe
ss
iona
l
b
od
ie
s
li
ke
IEEE,
A
CM,
CS
I,
et
c
.
C
urre
ntly
h
e
is
serving
as
a
R
ec
t
or
in
RRGroup
of
Instit
uti
ons
,
B
anga
lor
e.
He
h
a
s
serve
d
in
var
ious
ca
pa
ci
t
ie
s
as
Senior
Scie
n
ti
st,
Ind
ia
n
Spa
ce
Resea
rch
(
PA
RAM
Aw
ard
ee
),
Vice
Pr
eside
n
t
Resea
r
ch,
DS
Instit
uti
ons
,
B
anga
lor
e,
Visi
tin
g
Resea
rch
Profess
or,
Univer
sit
y
of
Ulster,
UK
.
Form
er
Saudi
Aram
co
Endow
ed
Chai
r
,
T
ec
hnol
og
y
and
Inform
at
i
on
Mana
gement, P
MU
.
KS
A
-
2015
trg
nair@gm
ai
l.
com
,
www.
tr
gnai
r.org
Evaluation Warning : The document was created with Spire.PDF for Python.