Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
6, N
o
. 4
,
A
ugu
st
2016
, pp
. 18
28
~
1
838
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
4.9
956
1
828
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Disaster Recovery Services
in Intercloud Using Genetic
Algorithm Load Balancer
T
a
ma
nn
a Jen
a
1
, J.
R.
Mohanty
2
1
School of Com
puter
Engineerin
g, KIIT University
2
School of Com
puter Applic
atio
n, KIIT University
Article Info
A
B
STRAC
T
Article histo
r
y:
Received
Ja
n 19, 2016
Rev
i
sed
Jun
23,
201
6
Accepte
d
J
u
l 10, 2016
Paradigm need
to shifts from cloud co
mputing
to intercloud
f
o
r disaster
recover
i
es
, whi
c
h can outbr
ea
k an
ytim
e and
an
y
w
her
e
. Na
t
u
ral dis
a
s
t
er
treatment includ
es radica
lly
hig
h
voluminous i
m
pa
tient job request which
dem
a
nds im
m
e
diat
e att
e
nt
ion.
Under the di
sequilibrium
ci
rcum
stance
,
inter
c
loud is mo
re practical and
functio
n
a
l optio
n. There are need of set of
protocols lik
e
quality
of serv
ices, se
rvi
ce l
e
v
e
l agr
eem
ent a
nd dis
a
s
t
er
recover
y
pacts to be disc
ussed
and clarif
ied du
ring the
initial setup to f
a
st
track
the
dis
t
r
e
s
s
s
cenario
.
Orches
trat
ion o
f
res
ources
in
larg
e s
c
a
l
e
distributed s
y
stem having muli-objec
tive o
p
timization of
resources,
minimum energ
y
consumption, ma
ximum throughput, lo
ad balan
c
ing
,
minimum carbon footprint altog
e
ther is
quite
ch
allenging. Intercloud where
resources of diff
erent
clouds are in a
lign
and play
s crucia
l role
in mapping
capab
ili
t
y
. Th
e objec
tive of thi
s
paper
is to im
provise and fast track th
e
mapping proced
ures in cloud platform
which can addr
ess impatient jo
b
requests in
balanced
and
efficient ma
nner. Genetic
algor
ithm based resource
allo
cation is pro
posed using par
e
to
optimal mapping of resour
ces to k
eep
high utilization
rate of pr
o
cessors, high
througp
ut and
low carb
on footprin
t.
Decision v
a
riables include utilization
of pro
cessors, throughput,
locality
cos
t
and real t
i
m
e
de
adlin
e. Sim
u
lati
on results
of load balancer using first in first
out and
gen
e
tic
algorithm ar
e
co
mp
ared under
similar circumstan
ces.
Keyword:
C
l
ou
d c
o
m
put
i
n
g
Gen
e
tic al
g
o
rith
m
Gree
n cl
ou
d c
o
m
put
i
ng
Intercl
o
u
d
Loa
d
bal
a
nci
n
g
Copyright ©
201
6 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Tam
a
nna Jena
,
Sch
ool
o
f
C
o
m
put
e
r
E
ngi
neer
i
ng,
KI
IT Uni
v
er
sity
,
B
h
u
b
a
n
eswa
r,
Odi
s
ha,
I
ndi
a.
Em
a
il: ta
man
n
a
sin
ghd
eo@gmail.co
m
1.
INTRODUCTION
C
l
o
u
d
C
o
m
put
i
n
g
i
s
a
hi
g
h
l
y
i
n
t
e
grat
ed
a
n
d
fl
exi
b
l
e
c
o
m
put
i
ng ser
v
i
c
e w
h
ere
use
r
n
eed
not
o
w
n
infra
struct
ure
,
serve
r
,
datace
nter etc. Using
a de
vice lik
e s
m
art phone, ta
blet, PC et
c. en
ab
led
with
intern
et,
provide access to c
o
m
putability of clou
d provi
der
while paying for its usag
e. Cl
oud c
o
m
puting ga
ve
us the
room
for m
obility
to access data from
anywhe
re unlike t
r
aditional setups whe
r
e we
need to be i
n
the sa
me
location for da
ta
availability.
Datacente
rs basically
us
e hypervisor ba
se
d
virtualiza
tion to exec
ute service.
Clo
u
d
Co
m
p
u
t
in
g
is b
a
sically a n
e
w m
e
tap
h
o
r
fo
r
sub
s
cr
iptio
n
b
a
sed
u
tili
ty
serv
ice
lik
e electricity,
tele
co
m
,
i
n
t
e
rnet
et
c.
It
s effect
on
h
u
m
an l
i
f
e and e
nvi
ro
nm
ent
i
s
di
scuss
e
d i
n
m
a
ny
co
nt
ra
di
ct
ory
way
s
. Th
e
r
e are
three m
odels
of cloud service
s
: Infrast
ruct
ure as a se
rv
ice
(IaaS), Soft
ware as a se
rvice
(SaaS), and
Platform
as a service (PaaS). Google
Apps and Salesforce.c
o
m
ar
e of type Saa
S
, where user
uses the applications
avai
l
a
bl
e wi
t
h
out
a
n
y
cont
r
o
l
o
v
er t
h
e
h
o
st
. I
n
PaaS
, pl
at
fo
rm
i
s
provi
ded
w
h
i
c
h
i
s
m
o
re or l
e
ss an
appl
i
cat
i
o
n f
r
a
m
ewor
k.
Exa
m
pl
es usi
ng
P
aaS are
G
o
ogl
e
App Engine, Am
azon
Web Services (AWS) etc.
User
u
s
es com
p
u
t
ab
ility
o
f
resou
r
ces, st
o
r
ag
e,
p
r
o
cessin
g
p
o
wer etc. in
IaaS, ex
am
p
l
e are Eu
calyp
tu
s.
Precisely use
r
/clients can lea
s
e com
puting resources
from
IaaS,
use a
p
plications
from
SaaS and ca
n
use as
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Disaster Recov
e
ry
Services
in Intercloud Usi
n
g
Gen
e
tic Algo
rithm Loa
d
Ba
lan
cer (Ta
m
an
na
Jena
)
1
829
wel
l
as devel
o
p ap
pl
i
cat
i
ons
on t
h
e
pl
at
fo
r
m
pro
v
i
d
e
d
at
PaaS cl
ou
d.
C
l
ou
d has at
t
r
act
ed h
uge
vol
um
e of
u
s
ers
b
ecau
s
e o
f
its elastic, flex
ib
le, con
n
ectio
n
l
ess,
h
i
g
h
co
m
p
u
t
ab
ility, in
v
e
stm
e
n
t
free, u
s
er friendly, an
d
secu
red
st
o
r
age cap
acity serv
ices.
Users
do
no
t g
e
t t
h
e tran
sp
aren
cy
o
f
its ex
ecu
tion
u
n
til and
un
less it is
ex
p
licitly d
e
man
d
e
d
as per
serv
ice lev
e
l ag
reem
en
t (S
LA).
Perform
a
n
ce o
f
cloud
dep
e
nd
s
h
i
gh
ly o
n
th
e
av
ailab
ility an
d
streng
t
h
o
f
i
n
tern
et. Slow
co
mm
u
n
i
ca
tio
n
in
creases waitin
g
,
thu
s
in
creasin
g
m
a
k
e
span
of
application altoget
h
er. Like a
ll technol
ogy “cloud” has its share
of pros and
cons. It is noticed that because of
huge c
o
mm
erc
i
al aspect, large num
b
er
of public a
n
d private clouds a
r
e
cr
eated i
n
last half
deca
de,
but
i
n
m
o
st o
f
th
e cases u
tilizatio
n
is less th
an
5
0
%. Leap
in
p
e
rfo
rman
ce is o
f
ten
p
r
o
c
laim
ed
w
h
ereas rise in
carbon
foo
t
prin
t is ig
no
red
.
State o
f
art o
f
ene
r
gy efficiency on large scale di
stri
bute
d
syste
m
a
nd its causes a
r
e in
literatu
re
o
f
m
a
n
y
research
pap
e
rs [1
],[2
].
Carbo
n
em
issi
o
n
cau
sed
du
e
to
ov
er pro
v
i
si
o
n
i
n
g
, serv
er sp
oo
ling
and c
ooli
ng
da
tacenters is immensely
contri
buting towards
global
warm
ing.
Its net
w
orking infrastruc
ture is
q
u
ite stron
g
,
with
h
i
gh
po
ten
t
ial fo
r a b
e
tter
en
v
i
ron
m
en
t with
lesser carbo
n
em
issio
n
,
if do
n
e
righ
t from
i
t
s
i
n
fa
ncy
or a
d
opt
i
o
n o
f
ne
w
e
r t
echni
que
s l
i
k
e serve
r
co
nsol
i
d
at
i
o
n
,
t
h
erm
a
l
m
a
nagem
e
nt
, opt
i
m
al l
o
ad
bal
a
nci
n
g et
c.
No
wa
day
s
al
m
o
st
al
l
busi
n
esses a
r
e m
ovi
ng t
o
cl
oud to re
duce their cost
of i
n
f
r
ast
r
uct
u
re,
main
ten
a
n
ce,
car
bon
f
o
o
t
pr
in
t, staf
f
i
ng
[3
]. I
n
or
d
e
r to keep the carbon em
ission low, trade off
betwe
e
n
crucial feat
ure
s
like privacy
, pe
rfor
m
a
nce and exte
nt of virt
ualizati
o
n
are ev
alu
a
ted
iterativ
ely.
Clo
ud
fed
e
ratio
n is an
in
teg
r
ated clo
u
d
st
ru
cture
wh
ich
co
m
e
s in
to
ex
isten
ce
with
b
e
tter fault to
leran
ce and
clean
er
wo
rl
d
co
nce
p
t
.
It
pot
e
n
t
i
a
l
t
o
re
duce
di
gi
t
a
l
wast
e a
n
d ca
r
b
o
n
em
i
ssi
on i
s
h
u
g
e. C
l
ou
d
com
put
i
n
g m
odel
,
whe
r
e num
e
rous user
utilizes single cloud
datacentre stru
ggle seve
ral cha
llenges.
In
order to inc
o
rpora
t
e ECC
(El
a
st
i
c
C
l
oud
C
o
m
put
i
ng)
,
m
o
st
cl
oud p
r
ovi
der est
a
bl
i
s
h l
a
rge
dat
ace
nt
er. T
h
ey
so
m
e
t
i
m
e
reserv
e l
a
rge
cap
acities to
inco
rpo
r
ate su
dden
vo
lu
m
i
n
o
u
s
j
o
b
requ
ests,
wh
ich
resu
lt un
d
e
ru
tilized
reso
urces i
n
m
o
st cases.
C
a
rb
on em
i
ssions
of l
a
r
g
e se
t
up co
nt
ri
but
e
t
o
wa
rds t
ech
ni
cal waste and
energy wa
ste.
These above factors
co
m
b
in
ed
ly con
t
ribu
te toward
s g
l
o
b
a
l
warmin
g
.
It is an o
p
e
n
q
u
e
stion
fo
r all hu
man
ities. Evo
l
u
t
io
n
of
m
echani
c
s an
d
t
echnol
ogy
ha
s t
a
ken a bi
g t
o
l
l
on nat
u
re as
a resul
t
we are
enco
u
n
t
e
ri
n
g
n
a
t
u
ral
di
sast
er
m
o
re
freq
u
e
n
tly th
an
ev
er. Ab
rup
t
nu
m
b
er o
f
job
requ
ests
h
a
v
i
n
g
d
i
fferen
t
j
o
b
lo
cality can
b
e
raised
at the sam
e
t
i
m
e
whi
c
h m
a
y
or m
a
y no
t
be i
n
t
e
rde
p
e
nde
nt
(c
om
pl
exi
t
y
i
n
creases
m
o
re t
h
an t
w
o
fol
d
s i
n
c
a
se of
in
terd
ep
en
d
e
n
t
i
m
p
a
tien
t
j
o
b
requ
ests). Ag
en
cies lik
e gov
ern
m
en
t, mili
tary, h
ealth
, security can
b
e
im
p
a
tien
t
at ti
mes. From user pe
rspective sing
le cloud datacente
r re
liability is
inad
equate. For a
reliable, flexi
b
le and
gree
ner
ap
p
r
oa
ch i
n
t
e
grat
e
d
c
l
ou
ds i
s
a
sm
art
e
r ene
r
gy
e
ffi
c
i
ent
o
p
t
i
o
n
.
1.
1.
Intercloud
and its c
o
m
p
one
n
ts:
In
tercl
o
ud
is an
e-i
n
teg
r
ated
in
frastru
ct
u
r
e
h
a
v
i
n
g
ex
ten
d
ed
co
m
p
u
t
ation
a
l cap
ab
ility
for u
tilization
of
resources and storage ca
pa
city. Its
m
odel is like
a sea
m
l
e
ss functionality
m
e
sh
where
each node is a cloud
or its com
pone
nt. In othe
r words, z
o
om
ed
out m
u
ltiple
cloud
net is intercloud.
The int
e
roperating
betwe
e
n
m
u
l
tip
le clo
u
d
s n
eed
to
b
e
in
co
m
p
lian
ce a
m
o
n
g
its
m
e
m
b
er. It is a powe
r
ful platf
o
rm
, can be real re
w
a
rdi
n
g
whe
n
its strength a
r
e e
x
ercis
e
d e
fficiently.
Here
job
re
quest is a servic
e to
be e
x
ecut
e
d
by cloud for
any
user/
e
nt
er
pri
s
e.
Job
re
quest
s
are m
a
i
n
l
y
divi
de
d i
n
t
o
sm
al
l
t
a
sks. Eac
h
t
a
sk ca
n be
execut
e
d by
one
o
r
m
u
l
tip
le clo
u
d
. Ob
j
e
ctiv
e is
o
p
tim
u
m
u
tiliz
atio
n
o
f
re
so
urces in
accordan
ce
with
SLA,
wh
ich
ap
paren
tly
l
e
ssens ca
rb
o
n
em
i
ssi
on. J
o
b
r
e
que
st
s are
di
v
i
ded i
n
t
o
m
a
ny
cat
ego
r
i
e
s l
i
k
e
nat
u
re
of s
e
r
v
i
ce (Iaa
S
, Pa
aS
an
d
SaaS), size, com
p
lexity, loc
a
tion et
c. Jo
b-
req
u
est
s
are s
o
m
e
tim
es di
vided i
n
t
o
C
P
U
i
n
t
e
nsi
v
e t
a
sks a
n
d
in
pu
t/o
u
t
p
u
t
i
n
ten
s
iv
e task
s.
Research
on
interclo
ud
arch
it
ecture a
n
d agre
e
m
ents is a virgin t
opic, neither wel
l
defi
ned
n
o
r
un
derst
o
o
d
e
n
o
u
g
h
.
I
n
o
u
r
pa
pe
r arc
h
i
t
ect
ure
o
f
i
n
t
e
rcl
o
u
d
i
s
exp
r
esse
d i
n
fi
gu
re
1. It
e
v
en
suf
f
er
s
fr
om
t
e
r
m
i
nol
ogy
am
bi
gui
t
y
.
A st
at
e
of a
r
t
o
f
i
n
t
e
r
-
cl
o
u
d
i
s
do
ne t
o
a
n
e
x
t
e
nt
i
n
[
4
]
-
[
7
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
JECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
18
28
–
1
838
1
830
Fi
gu
re 1.
A
r
chi
t
ect
ure of I
n
t
e
r
c
l
o
u
d
Fi
gu
re 2.
G
A
B
a
sed
l
o
a
d
bal
a
ncer
C
o
m
pone
nt
s o
f
Int
e
rcl
o
u
d
:
Intercl
o
u
d
M
a
nage
r
Intercl
o
u
d
A
g
r
e
em
ent (SL
A
a
g
reem
en
t between
peer cl
ou
d p
r
ov
id
ers)
Inter Cloud Resource Ma
na
ge
r (health
, availability and
usa
g
e
of each re
source
is audited
peri
odically)
Reputation M
a
nage
r
(pe
r
form
ance of
each com
pone
nt a
nd
fee
dbac
k
from
user is docum
ented, whi
c
h
reflects in its rep
u
t
ation
i
n
mark
et
)
Mo
tiv
atio
n fo
r
In
tercl
o
ud
:
It is fu
tile and in
co
m
p
eten
t to
h
a
v
e
a si
n
g
l
e clo
u
d
p
r
o
v
i
der d
a
tacen
t
er
worldwid
e t
o
co
m
b
at d
i
saster
reco
very
.
Peer
pr
o
v
i
d
e
r
s
nee
d
t
o
be
wel
l
i
n
t
e
g
r
at
ed
fo
r
com
m
on goal
,
pr
o
v
i
d
i
n
g
Q
o
S an
d
wo
rl
d
w
i
d
e
com
p
l
i
a
nce wi
t
h
di
st
ri
b
u
t
e
d r
e
so
urces
. M
ovi
ng t
o
i
n
t
e
rcl
o
u
d
i
s
t
h
e next
b
e
st
l
ogi
cal
opt
i
on
fo
r bet
t
e
rm
en
t
of
bot
h pr
o
v
i
d
ers an
d use
r
s p
e
rspect
i
v
e i
n
o
r
der t
o
a
voi
d se
rve
r
s p
o
o
l
i
n
g
,
red
u
ce car
b
o
n
-
fo
ot
p
r
i
n
t
et
c. To
put c
h
ec
k
on c
a
rbon foot-print and
ene
r
gy waste by
optim
al utilization
of e
ach c
o
m
pone
nt
of intercl
o
ud.
It
i
s
hi
gh t
i
m
e
t
o
i
n
cor
p
orat
e
gree
n net
w
or
ki
n
g
, com
put
at
i
on an
d com
m
uni
cat
i
o
n. C
o
n
t
ri
but
i
o
n
o
f
wireless tech
no
log
y
and
its i
m
p
act on
en
v
i
ron
m
en
t is
d
e
tailed
i
n
literatu
re [8
]-[10
]
. Th
e
In
ternatio
n
a
l
Tel
ecom
m
uni
cat
i
on
Uni
o
n
T
e
l
ecom
m
uni
cati
ons
O
p
erat
i
o
n
s
(I
TU
-T)
o
p
e
r
at
es on
m
oni
t
o
ri
n
g
st
an
da
rd
on
a
n
in
tern
ation
a
l b
a
sis fo
r fair an
d
co
m
p
etitiv
e
m
a
rk
et. Th
ere is align
m
en
t b
e
tween
teleco
mm
u
n
i
catio
n
depa
rt
m
e
nt
and en
vi
r
onm
ent
a
l
i
ssues cause
d by
t
h
em
. It
i
s
fo
u
nd t
h
at
ne
t
w
o
r
ki
ng
wast
e i
s
subst
a
nt
i
a
l
l
y
hi
gh
at
end u
s
er si
t
e
. Thi
s
ca
n be
add
r
esse
d by
o
p
t
i
ng
wi
re
d i
n
t
e
rnet
at
use
r
l
e
vel
or
di
rect
i
o
nal
ant
e
n
n
as
. Whe
n
eq
u
ilibriu
m
is
attain
ed
b
e
tween
jo
b-request an
d
res
ources th
en
p
r
essu
re sh
ifts t
o
n
e
twork
i
n
g
. Faster
net
w
or
ki
n
g
p
r
ot
oc
ol
w
h
i
c
h i
s
i
n
t
r
o
d
u
ced i
n
2
0
0
0
, cal
l
e
d
“Infi
n
i
B
an
d (
I
B
)
” i
s
co
nsi
d
ered as a
n
ad
vanc
e
d
en
erg
y
efficient altern
ativ
e to trad
ition
a
l TC
P/IP.
It a
llows
read
i
n
g and
writin
g
d
a
ta
fro
m
rem
o
te co
m
p
u
t
er’s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Disaster Recov
e
ry
Services
in Intercloud Usi
n
g
Gen
e
tic Algo
rithm Loa
d
Ba
lan
cer (Ta
m
an
na
Jena
)
1
831
m
e
m
o
ry
by
ef
fect
i
v
el
y
by
p
a
ssi
ng
t
h
e
o
p
e
r
at
i
n
g
sy
st
em
[1
1]
.
Apa
r
t
f
r
o
m
i
t
s
huge
p
o
t
e
nt
i
a
l
i
t
i
s
n
o
t
u
s
ed
m
u
ch i
n
Had
o
op
or cl
ou
d
p
l
at
form
as i
t
suf
f
ers
fr
om
t
h
e
m
i
sconcept
i
on
of
bei
n
g e
xpe
nsi
v
e. I
n
t
e
rcl
o
ud
services are di
vide
d into thre
e com
p
lim
e
ntary com
ponent
s
discussing the
integratio
n and inter-opera
b
i
lity of
m
u
l
tip
le layered
Cloud
Mod
e
l Dev
i
ce: 1] In
tercl
o
ud
fed
e
ration
2
]
In
tercloud
co
n
t
ro
l and
co
m
p
lian
ce
m
a
nagem
e
nt 3] Intercl
o
ud m
u
lti layer interface am
ong
Iaa
S
, Paa
S
a
n
d Sa
aS [12
]
.
In
tercl
o
ud
shou
ld
satisfy
fo
llo
wi
n
g
go
als:
C
o
m
pone
nt
s o
f
i
n
t
e
rcl
o
u
d
s
h
o
u
l
d
su
p
p
o
r
t
com
m
uni
cat
ion
bet
w
een
d
i
ffere
nt
ser
v
i
c
e l
a
y
e
r (ve
r
t
i
cal
in
teg
r
ation), d
i
fferen
t
do
m
a
in
s
and
app
licati
o
n
s
an
d serv
ices (h
or
izon
tal in
tegratio
n).
QoS (Qu
a
lity
o
f
Serv
ices) sh
ou
ld
b
e
m
a
in
tain
ed
acro
ss b
o
t
h
ho
rizon
t
al as well
as v
e
rtical in
teg
r
ation
al
on
g
wi
t
h
opt
i
m
i
zat
i
on of
co
m
put
i
ng an
d
st
ora
g
e.
Ru
le ou
t v
e
nd
or
lo
ck
-
i
n
:
Ven
dor
lo
ck
-
i
n
is an
unf
ortunate situ
atio
n
,
i
n
wh
ich u
s
er
can
no
t tran
sit its
servi
ce
fr
om
one p
r
ovi
der t
o
i
t
s
com
p
et
it
or.
It
i
s
fo
u
nd
o
u
t
t
o
be t
h
e m
a
jor i
m
pedim
e
nt
for e
n
t
e
r
p
ri
ses t
o
ad
op
t cloud
.
In
co
m
p
atib
le bu
sin
e
ss pro
p
rietary tech
no
log
i
es are t
h
e
hu
rd
les, as a resu
lt ad
ap
tab
ility of
pr
o
v
i
d
er
an
d
o
f
user
re
qui
rem
e
nt
are
n
o
t
i
n
c
o
m
p
l
i
a
nce.
EC2
(Elastic
Clo
u
d
Co
m
p
u
tin
g) is li
m
i
ted
wh
en
rest
ri
ct
ed t
o
si
n
g
l
e
p
r
o
v
i
d
e
r
. S
u
dde
n spi
k
es
of
us
er
req
u
i
r
em
ent
cann
o
t
be acc
om
m
odat
e
d by
an
y
si
ngl
e cl
ou
d pr
o
v
i
d
er i
n
al
l
t
i
m
e
s wi
t
hout
causi
n
g
pe
ri
l
o
us
po
we
r c
ons
um
pt
i
o
n
an
d e
x
po
nent
i
a
l
hi
g
h
c
o
st
.
In
ord
e
r to
wrap
all th
e attractiv
e
feat
u
r
es
of cl
ou
d,
i
t
i
s
i
n
t
h
e
best
i
n
t
e
rest
o
f
bot
h
p
r
o
v
i
d
e
r
a
n
d
cust
om
er t
o
h
a
ve a hi
g
h
st
anda
r
d
l
e
gi
sl
at
i
on
of cl
o
u
d
f
e
derat
i
o
n. P
r
ot
ocol
re
ga
rdi
n
g
QoS
,
de
pl
oy
m
e
nt
,
main
ten
a
n
ce,
resou
r
ce allo
cat
io
n
,
arb
itrag
e,
repu
tatio
n
qu
otien
t
, peeri
n
g
agreem
ent etc.
nee
d
t
o
be t
h
ou
g
h
t
t
h
r
o
u
g
h
m
i
nut
el
y
.
Int
e
rcl
o
ud
bei
n
g n
ovel
,
i
t
suf
f
ers
fr
om
am
bi
guou
s t
e
rm
i
nol
o
g
y
.
A
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
enc
e
t
ool
s
have
bee
n
use
d
as optim
ization sche
duler in
cloud platform
.
When sea
r
c
h
space is huge,
constraints are
m
u
lti
o
b
j
ectiv
e and
d
ead
lin
e is stri
ct, CI (C
o
m
p
u
t
atio
n
a
l In
te
llig
en
ce) t
o
o
l
s are h
i
gh
ly produ
ctiv
e. In
o
u
r
p
a
p
e
r
we
are usi
n
g t
h
e
paret
o
-
opt
i
m
al sol
u
t
i
o
ns co
n
cept
fo
r res
o
ur
ce al
l
o
cat
i
on t
echni
que i
n
w
h
i
c
h, eac
h sol
u
t
i
on i
s
ev
alu
a
ted
u
n
d
e
r m
u
ltip
le criteria. Sub
s
et of
op
tio
ns is
d
e
termin
ed
, su
ch
that th
ere is no
ro
o
m
o
f
im
p
r
o
v
e
m
en
t
in the
provided s
o
lutions
. Health of
eac
h
c
o
m
pone
nt is
evaluate
d i
t
eratively.
When
one
or m
u
ltiple
com
pone
nt (ba
s
ically server, datacente
r, internet,
network) of cloud cro
sses lowe
r threshold of fitne
ss by
natural/accidental/technical disaster,
the
n
re
-routing of resource
allocatio
n is nee
d
e
d
. These sce
n
ari
o
s are
recu
rre
nt
du
e t
o
m
a
ny reaso
n
s
such as o
v
er
pr
o
v
i
s
i
oni
ng
o
f
reso
u
r
ces, su
dde
n spi
k
e of i
m
pati
ent
job re
que
st
s,
l
o
w i
n
t
e
r
n
et
ba
nd
wi
dt
h,
hi
g
h
i
n
t
e
rnet
t
r
a
ffi
c
,
hi
g
h
ca
rb
o
n
f
oot
pri
n
t
,
hi
g
h
el
ect
ri
ci
t
y
cons
um
pt
i
on f
o
r
co
ol
i
n
g
datacenters,
power failure a
n
d natural calam
ity.
Loa
d
bal
a
nci
n
g i
s
one
of t
h
e
m
o
st
i
n
fl
uent
i
a
l
aspect
s of cl
ou
d com
put
i
n
g
,
i
t
can be br
oa
dl
y
di
vi
de
d
i
n
t
o
cent
r
al
i
z
e
d
o
r
dece
nt
ral
i
zed and
dy
na
m
i
c or st
at
i
c
. An
ot
he
r co
nc
ept
use
d
i
n
t
h
i
s
paper i
s
G
e
net
i
c
Algo
rith
m
(GA), a st
o
c
h
a
stic op
timizatio
n
to
o
l
u
s
i
n
g
Darwi
n
th
eo
ry
wh
ich
g
i
v
e
s
b
e
tter resu
lt
wh
en search
space is
ve
ry
high. All t
h
e
possible s
o
lutions are
term
ed
chrom
o
som
e
s; fitness
valu
e
of each chrom
o
som
e
is
ev
alu
a
ted
.
Selectio
n
p
r
o
c
ed
ure is
u
s
ed
to filter so
m
e
so
lu
tio
ns to th
e
n
e
x
t
g
e
n
e
ration
p
o
o
l
.
Op
erat
o
r
s
of
genet
i
c
al
g
o
ri
t
h
m
are used o
n
t
h
e ch
rom
o
som
e
s t
o
obt
ai
n opt
i
m
i
zed sch
e
dul
e. Eac
h
sc
hed
u
l
i
n
g st
rat
e
gy
has
i
t
s
st
rengt
h a
n
d we
ak
nesses
.
Im
pl
em
ent
a
t
i
on t
i
m
e i
s
ver
y
fast
an
d
gi
v
e
s bet
t
e
r
resul
t
whe
r
eas
pre
m
at
ure
co
nv
erg
e
n
ce is its do
wn
si
d
e
. Co
nsequ
e
n
tly lo
ad
b
a
lan
c
e
of each
serv
er an
d v
i
rtu
a
l m
a
c
h
in
e is m
a
in
tai
n
ed to
av
o
i
d
wo
rk
overlo
a
d
.
Virt
u
a
l
i
zatio
n
o
f
p
r
o
c
esso
r is u
s
ed
in
sim
u
lat
i
o
n
to en
cou
r
ag
e m
u
ltiten
a
n
c
y. Op
t
i
m
u
m
u
tilizatio
n
of
resources an
d equ
ilib
riu
m
b
e
tween
d
e
m
a
n
d
and
sup
p
l
y
can
m
a
k
e
in
t
e
rclou
d
as th
e m
o
st
b
e
n
e
ficial u
tility serv
ice of t
h
e era.Th
is p
a
p
e
r is
o
r
g
a
n
i
zed
as
fo
llows sectio
n
2
prov
ides related
research
on
cl
ou
d com
put
i
n
g
,
i
n
t
e
rcl
o
u
d
and sc
he
dul
i
n
g use
d
i
n
cl
o
u
d
pl
at
f
o
rm
. Sect
i
on 3 a
n
al
y
zes ou
r m
odel
usi
n
g
opt
i
m
i
zati
on t
echni
que
s cal
l
e
d G
A
base
d
l
o
ad
bal
a
nc
e
r
.
Sect
i
o
n 4 descri
bes resu
l
t
and di
scuss
i
on
o
n
sim
u
l
a
t
i
on.
Se
ct
i
on 5 di
scus
s
e
s
co
ncl
u
si
on
.
2.
RELATED RESEARCH
Arch
itectu
r
e, tech
n
i
q
u
e
s and
ro
le of co
m
p
onen
t
s o
f
cl
o
u
d
co
m
p
u
tin
g
is men
tio
n
e
d
in
th
e literatu
re o
f
m
a
ny
researc
h
pape
rs [
2
]
,
[
1
0]
,[
12]
,
[
1
3
]
.
P
o
w
e
r m
a
nagem
e
n
t
i
n
cl
ou
d pl
at
f
o
rm
i
s
one o
f
t
h
e i
m
pact
ful
t
opi
cs
.
Large c
h
unk of energy cons
um
ption is used in coo
ling datacenters. T
o
lessen th
e power usage in c
ooling
datacenters, load bala
ncer
plays vital
role. Nature of job-requests are ei
ther interde
p
endent
or inde
pe
nde
nt.
Int
e
r
d
e
p
en
de
n
c
y
bet
w
ee
n j
o
bs/
t
a
sks i
s
rep
r
esent
e
d
by
TI
G (Ta
s
k i
n
t
e
ra
ct
i
on
Gra
p
h
)
t
o
s
h
o
w
m
a
ppi
ng
o
f
di
ffe
re
nt
appl
i
cat
i
ons [
14]
. C
o
m
p
l
e
xi
t
y
of execut
i
o
n o
f
i
n
t
e
rde
p
e
nde
nt
t
a
sks bec
o
m
e
s
m
o
re t
h
a
n
t
w
o f
o
l
d
s in
case o
f
depe
n
d
ent
j
o
b
-re
q
u
e
s
t
s
. T
h
ey
ha
v
e
co
nsi
d
e
r
ed
het
e
r
oge
ne
ous
j
o
b
-re
q
u
est
s
whi
c
h a
r
e e
x
e
c
ut
ed
conc
urre
ntly and comm
unicating
with eac
h
othe
r at the
sa
me ti
me. Job-reque
sts f
aces queui
n
g in
getting the
req
u
i
r
e
d
reso
u
r
ces [
15]
.
Ha
ve
use
d
“R
eq
uest
depe
n
d
ent
st
rat
e
gy
” t
ech
ni
q
u
e
by
sc
he
dul
er
t
o
m
a
p j
o
b
-re
q
u
est
s
to
resou
r
ces.
Th
ey h
a
ve tried
to
estab
lish th
e relati
onsh
i
p bet
w
ee
n j
o
b arri
val
rat
e
, rat
e
of exec
u
t
i
on o
n
waitin
g–
tim
e o
f
job
-
requ
ests
an
d
rate of emp
tin
ess
o
f
q
u
eu
e. Th
e
d
r
awback
of requ
est
d
e
p
e
nd
en
t strat
e
g
y
is
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
18
28
–
1
838
1
832
if n
u
m
b
e
r o
f
inco
m
i
n
g
j
o
b
-
req
u
e
st d
ecreases th
en
its i
m
p
a
ct o
n
u
tilizatio
n
rate o
f
serv
ers is
m
o
re sev
e
re th
an
othe
r sche
dule
r
s. Since s
p
eci
fic serve
r
s are
allotted to cer
tain
typ
e
o
f
jobs so
m
e
t
i
m
es c
a
u
s
es serv
er sp
rawl.
Job
-
r
e
q
u
est
un
der
g
o que
ui
n
g
o
n
ce regi
st
ere
d
wi
t
h
cl
ou
d p
r
o
v
i
d
e
r
oe
i
n
t
e
rcl
o
ud
.
Ty
p
e
s and
t
r
e
nds
f
o
u
n
d
i
n
que
ui
n
g
ca
nn
o
t
be ove
rl
o
oke
d. Di
ffe
rent
fe
at
ures o
f
q
u
e
u
i
ng are
descri
b
e
d spa
n
ni
n
g
b
a
l
k
i
n
g
,
re
negi
n
g
an
d
joc
k
ey
i
n
g i
n
cl
ou
d
[
16]
.
Balk
in
g
is a situ
atio
n
wh
ere
cu
sto
m
er d
ecid
e
s no
t to
jo
in th
e q
u
e
u
e
if it is to
o
lo
ng
.
In
Ren
e
g
i
ng,
cust
om
er l
eav
es t
h
e
q
u
eue
a
f
t
e
r
havi
ng
wa
i
t
e
d f
o
r t
o
o l
o
ng
f
o
r
ser
v
i
ce.
Joc
k
ey
i
n
g
de
scri
bes
t
h
e e
x
t
e
nt
o
f
im
pat
i
e
nt
user
s, whe
r
e cu
st
o
m
er shi
f
t
s
bet
w
een
que
ues t
o
be ser
v
e
d
fa
st
. They
hav
e
t
r
i
e
d t
o
est
a
bl
i
s
h t
h
e
rel
a
t
i
ons
hi
p be
t
w
een us
er bal
k
i
n
g/
re
negi
ng
and sy
st
em
blocki
ng
on t
h
e
t
h
r
o
u
g
h
p
u
t
rat
e
s. Nat
u
re of
di
gi
t
a
l
users
is analyz
ed which im
pa
ct thr
oug
hpu
t an
d
u
s
ag
e
o
f
r
e
so
ur
ces.
In
[17
]
, th
ey p
r
op
osed an
algo
r
i
t
h
m
w
h
er
e
ban
d
w
i
d
t
h
a
r
e
share
d
bet
w
e
e
n VM
s
dy
na
m
i
cal
ly
, i
n
or
der t
o
m
a
p i
m
pati
ent
jo
b r
e
que
st
s by
M
i
ni
m
u
m
C
o
m
p
l
e
t
i
on Ti
m
e
(M
C
T
) schedul
i
n
g al
g
o
ri
t
h
m
.
Howe
ver
preci
se an
al
y
s
i
s
on
per
f
o
r
m
a
nce are fo
u
nd m
i
ssi
ng.
Clu
s
tering
job
-
requ
ests in
acco
r
d
a
n
ce to
reso
urce
p
r
o
ce
ssi
n
g
cap
a
b
ility i
s
d
o
n
e
in
[1
8
]
. Th
ey h
a
v
e
sug
g
e
sted
j
ob-grouping a
l
gorithm
in which
us
e
r
tasks are team
ed according t
o
resource processing
ca
pa
bility a
nd
gr
o
upe
d t
a
s
k
s
are assi
g
n
ed
t
o
di
ffe
re
nt
reso
u
r
ces i
n
c
l
ou
d c
o
m
put
i
ng.
Si
nce
real
l
i
f
e ex
peri
m
e
nt
s of
i
n
t
e
rcl
o
ud a
r
e
very
e
xpe
nsi
v
e
and
di
f
f
i
c
ul
t
t
o
i
m
pl
em
ent
,
oft
e
n si
m
u
l
a
t
i
o
ns are
wi
ser
o
p
t
i
on t
o
e
nha
n
ce t
h
e
per
f
o
r
m
a
nce.
Op
tim
u
m
task
co
nso
lid
ation
u
s
ing
GA
for
b
o
t
h
lon
g
term an
d
sho
r
t term is p
r
op
osed in
[1
9
]
, sh
e
su
gg
ested
that
resource
u
tiliz
atio
n
d
i
rectly relates to
en
ergy co
nsu
m
p
tio
n. Pro
cess
o
f
in
t
e
rclou
d
is m
i
mick
ed
in [20] using Si
m
I
C tool kit and Cl
oudSi
m
si
m
u
la
tor. It
is co
m
p
arativ
ely easy to set
up m
u
ltiple datacenter,
n
e
two
r
k
i
ng
ai
ds, stor
ag
e i
n
si
m
u
la
to
r
s
. I
t
is
f
oun
d
th
at
resu
lts of
Sim
I
C co
m
p
li
m
e
n
t
s Clo
u
d
S
im
. Scen
ari
o
s
whi
c
h sh
o
w
s
fai
l
u
re
basi
cal
l
y
needs ef
fe
ct
i
v
e dy
nam
i
c
m
a
ppi
n
g
t
echni
que
. A m
a
t
h
em
at
i
cal
m
o
del
i
s
prese
n
ted, whe
r
e virtual m
achine pl
acem
ent
is optim
a
lly allo
cated whe
n
ever bandwidt
h
capacity is available
an
d n
e
t
w
ork
l
a
ten
c
y is low
[
2
1
]
. Th
ey
h
a
v
e
u
s
ed
p
e
rf
orman
ce-
or
ien
t
ed
V
M
allo
catio
n appr
o
a
ch
i
n
cloud
com
put
i
ng
pl
at
fo
rm
. Sim
i
l
a
r
pape
r ba
sed
o
n
per
f
o
r
m
a
nce whe
r
e f
o
c
u
s i
s
on l
a
t
e
ncy
are
[2
2]
, t
h
ey
f
o
u
nd t
h
at
trans
f
erring
da
ta betwee
n i
n
tra-cloud provi
ders a
r
e m
o
re
cost effective than inter-c
l
o
ud provides
(datacenters
of sam
e
cl
oud pr
o
v
i
d
er
but
di
ffe
rent
ge
o
g
ra
phi
cal
l
y
di
st
ri
b
u
t
e
d l
o
cat
i
o
n)
. They
sug
g
est
e
d l
a
rge
r
R
I
p
u
r
c
hase
th
an
requ
ired
is b
e
tter in
mo
st cases. Alth
ou
gh
buying larger capa
c
ity can c
ontrib
u
te towa
rds s
e
rv
e
r
spo
o
l
i
n
g. M
a
n
y
pape
rs
have
use
d
C
I
t
o
ol
s t
o
m
a
p jo
b-
re
q
u
est
t
o
reso
u
r
c
e
s. Ge
net
i
c
al
g
o
ri
t
h
m
i
s
used
as l
o
ad
bal
a
nce
r
i
n
[2
3]
, whe
r
e si
ze of n
odes
req
u
i
red are deci
de
d de
pen
d
i
n
g o
n
t
h
e wo
r
k
l
o
a
d
. Si
m
u
l
a
t
i
on resul
t
s
t
o
wa
rds
cl
o
u
d
no
de m
i
nim
i
zat
i
on s
h
ows
n
u
m
ber of
n
ode
s
avai
l
a
bl
e i
s
t
w
i
ce t
h
e
wo
r
k
l
o
a
d
.
Adv
a
n
t
ag
es o
f
lo
ad
b
a
lan
c
ing
is
o
u
tlin
ed
wh
ereas
s
cen
ario
map
p
e
d
is no
t ap
pro
p
riate. Similarly
[2
4
]
have
use
d
f
u
z
z
y
NN sc
hed
u
l
e
r an
d B
e
r
g
er
M
odel
sche
d
u
l
e
r t
o
c
o
m
p
are t
h
e pe
rf
or
m
a
nces of
di
f
f
ere
n
t
sch
e
d
u
l
ers. Con
s
train
t
s con
s
id
ered
in
sim
u
latio
n
ar
e co
mp
letio
n
tim
e a
n
d
b
a
n
d
wid
t
h
u
tilizatio
n
.
Clou
d
s
im
si
m
u
lato
r is u
s
ed
to
m
a
p
task
s to
resou
r
ces; map
p
i
ng
ti
m
e
an
d
m
a
k
e
sp
an
ti
m
e
are
th
e p
e
rfo
r
m
a
n
ce
m
e
t
r
ics in
[25
]
. Th
ey h
a
v
e
in
clud
ed
time tak
e
n
to
st
ag
e in
th
e in
pu
t files an
d
stag
e ou
t th
e o
u
t
pu
t files to
d
e
si
red
l
o
cat
i
on. L
o
a
d
bal
a
nci
n
g i
n
cl
ou
d com
put
i
ng
do
ne
by
va
ri
o
u
s al
g
o
ri
t
h
m
,
t
a
bul
ar re
pr
esent
a
t
i
on
of
r
e
search
do
ne
o
n
l
o
a
d
b
a
l
a
nci
n
g
u
s
i
n
g
vari
ous
e
v
ol
u
t
i
onary
a
n
d s
w
arm
based al
g
o
r
i
t
h
m
i
s
sum
m
ari
zed i
n
[
2
6]
.
C
l
ou
d
users
are cate
g
orized
on the basis
of
jo
b-leng
th
, n
a
ture
o
f
u
s
er (do
m
estic o
r
co
mmercial) and
ti
me o
f
execut
i
o
n
(pea
k o
r
o
f
f
-
pe
ak
)
.
Di
f
f
ere
n
t
p
r
i
c
i
ng st
rat
e
gy
i
s
used t
o
p
r
om
ot
e off
-
pea
k
j
o
b an
d pe
nal
i
z
e
vol
um
i
nous
j
o
b-
req
u
est
at
pe
ak
ho
u
r
s
[2
7]
.
C
onsci
ou
s c
o
n
t
ri
but
i
o
n
fr
om
bot
h i
.
e
,
user
a
n
d
p
r
ovi
der
ca
n m
a
ke
diffe
re
nce in green cloud com
puting.
VM
m
i
grat
i
o
n
i
s
su
g
g
est
e
d
i
n
[2
8]
t
o
re
du
ce w
o
r
k
l
o
a
d
o
f
ove
rl
oa
de
d
n
ode
.
Depl
oy
m
e
nt
o
f
VM
mig
r
atio
n
cau
s
es d
a
ta
ov
erh
e
ad
an
d secu
tity syste
m
is att
ack
p
r
on
e.
A t
w
o layered clu
s
ter as a service is
propose
d
as disaster recovery services. First laye
r
m
a
nages t
h
e o
p
erat
i
ng sy
st
em
of t
h
e phy
si
cal
no
des
fo
rm
i
ng t
h
e cl
ust
e
rs
. Sec
o
nd
l
a
y
e
r, de
pl
oy
s
t
h
e s
o
ft
wa
re c
o
m
ponent
s t
o
i
n
st
al
l
on t
h
e
no
des
Di
sast
er
re
cove
ry
in
in
tercloud
syste
m
b
y
in
teg
r
atin
g
cl
u
s
ter as a serv
ice and is d
o
n
e
b
y
u
s
i
n
g
con
tin
ou
s-time Mark
ov
chain
s
[29].
Bac
k
up of
all
data of each no
de
of
clusters a
r
e sa
ved in e
n
c
r
ypt
e
d
form
with diffe
re
nt organization
placed i
n
distant ge
ogra
phical locati
on. Single bac
k
up see
m
s
inade
quate
and diffe
rent provider
m
a
y
caus
e
co
nfid
en
tiality
issu
es. Lo
ad
balan
cer
b
a
sed
o
n
GA is u
s
ed
in
[3
0
]
,
wh
ere
clo
u
d
a
n
a
lyst sim
u
la
to
r is u
s
ed. Th
ey
have c
o
nsi
d
e
r
e
d
si
n
g
l
e
cent
r
a
l
i
zed cl
ou
d dat
a
cent
e
r
wh
ich is stil
l a far-away d
r
eam
. Differen
t
tim
e-zo
n
e
and
d
i
fferen
t
d
a
ta-lo
cality co
st are no
t tak
e
n
i
n
to
co
nsid
eration.
3.
MO
DEL DE
S
CRI
PTIO
N
Each
jo
b
r
e
quest
co
m
p
r
i
ses o
f
gr
oup
o
f
inp
u
t
f
iles, gr
oup
o
f
ou
t
p
u
t
f
iles, Q
o
S r
e
qu
ire
m
en
t,
job
-
length size
(express
ed in M
I
PS), a
n
d arrival tim
e
and
da
ta
location.
In
our sim
u
lation each job-re
que
st is
d
i
v
i
d
e
d
in
to
task
s lik
e co
m
p
ute in
ten
s
iv
e and
I/O in
tensiv
e task
s. Each
task
is allo
cated
to
a v
i
rtu
a
l
m
a
ch
in
e.
Virtual m
achines are t
h
e
basic com
putational
units
of
datacenters.
We ha
ve c
o
ns
idere
d
10-15
physical
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Disaster Recov
e
ry
Services
in Intercloud Usi
n
g
Gen
e
tic Algo
rithm Loa
d
Ba
lan
cer (Ta
m
an
na
Jena
)
1
833
machines, a
nd
each physical
machine is capable of exte
ndi
ng
10 vi
rtua
l
machines. Let
D be t
h
e set of cloud
p
r
ov
id
ers/d
a
tacen
ters in
wh
ich
D= {d
1
, d
2
,…,
d
|D
|
}. Let J
b
e
th
e set o
f
job
-
requ
ests wh
i
c
h
are reg
i
stered
with
th
e clou
d con
t
ro
ller, eith
er
waitin
g
in
qu
eu
e
o
r
b
e
ing
ex
ecuted
, su
ch th
at J= {j
1
, j
2,
…
,
j
|n
|}
.
Let T
b
e
set
o
f
task
s
suc
h
that, T= {
(
t
11
, t
12, …,
t
1m
)
,
(
t
21,
t
22,
...,
t
2m
.)
,…,(t
n1,
t
n2,
…,
t
nm
)
},
whe
r
e m
repre
s
ent
s
n
u
m
b
er o
f
t
a
sks a
j
o
b-
re
que
s
t
i
s
di
vi
ded
.
In
o
u
r
si
m
u
l
a
t
i
on we
have
t
a
ke
n
m
ranges
fr
om
0 t
o
5.
Let
Qo
S (t
)
b
e
t
h
e
set
o
f
Qo
S
req
u
i
r
em
ent
fo
r task
, Fin
t
is th
e set of
input f
iles f
o
r
task
t an
d
Fou
t
t
is th
e set of
o
u
t
p
u
t files for task
t. Each
task
has two
d
ead
lin
es,
n
a
mely sd
t (start
d
e
ad
lin
e) and
fd
t
(fin
a
l
d
e
ad
lin
e). Th
e m
a
j
o
r obj
ectiv
e
o
f
th
is
work is to
op
timize
t
h
e
m
a
ppi
n
g
p
r
oces
s i
n
suc
h
a way
t
h
at
m
a
xi
m
u
m
num
ber of
jo
b re
ques
t
s get
s
execut
e
d wi
t
h
l
east
po
ssi
bl
e
reso
u
r
ces wi
t
h
out
com
p
r
o
m
i
si
ng SL
A co
nst
r
ai
nt
s. O
b
ject
i
v
e f
unct
i
o
n o
f
GA i
s
m
a
xim
i
zi
ng t
h
e t
h
ro
u
g
h
p
u
t
,
m
i
nim
i
zi
ng t
h
e num
ber o
f
phy
si
cal
res
o
u
r
ces an
d m
i
nim
i
zi
ng t
h
e
m
a
kes
p
an t
i
m
e.
Thr
o
ug
h
put
=
Ʃ
T
i,
T is
num
ber
of
j
o
b
req
u
est
s
w
hose
exec
ut
i
o
n
i
s
c
o
m
p
l
e
t
e
d bef
o
r
e
dea
d
l
i
n
e.
Obj
ectiv
e
fun
c
tion
=
max
_
t
hroug
hpu
t, m
i
n
_
d
atalo
cality
co
st,
m
i
n
_
m
ak
esp
a
n
,
min_physical_resources
Sin
ce M
O
GA (Mu
lti Obj
e
ctiv
e Gen
e
tic
Algo
rith
m
)
is
u
s
ed
b
a
sed
o
n
th
e pro
c
ess
of n
a
tural
ev
o
l
u
tio
n, ti
m
e
sp
en
t on
liv
e mig
r
atio
n
of task
s is n
o
t
co
nsi
d
ere
d
wi
t
h
i
n
t
h
e scope o
f
t
h
e pape
r. E
n
er
gy
l
o
ss i
s
i
d
ent
i
f
i
e
d as e
n
er
gy
w
h
i
c
h i
s
not
co
ns
um
ed by
any
sub sy
st
em
and ove
r
h
ead
of t
h
e s
u
pp
o
r
t
i
ng s
u
b
s
y
s
t
e
m
s
[2]. C
o
m
pone
nts of our m
odel is illustrated i
n
Fi
gure
2.
Model c
o
nstitutes of following c
o
m
pone
nts:
C
l
ou
d re
so
urce
sche
d
u
l
i
n
g
co
nt
r
o
l
l
e
r
C
l
ou
d Fe
derat
i
on
C
o
nt
r
o
l
l
e
r
Intra
fitnes
s ca
lculator
of eac
h
resource
Mu
lti-o
b
j
ective Gen
e
tic Algorith
m
lo
ad
b
a
lan
cer
Cloud c
ontroll
er (which
recei
ves
t
h
e job request)
Vi
rt
ual
res
o
u
r
c
e
p
ool
C
onst
r
ai
nt
s of
m
odel
:
Each
j
o
b
re
qu
e
s
t
nee
d
t
o
be
e
x
ecut
e
d
by
de
a
d
lin
e,
o
t
h
e
rwise it is sup
p
o
s
ed to
b
e
failed
.
Each
jo
b req
u
est can
b
e
allo
cated
to
m
o
re t
h
an
o
n
e
clou
d.
Im
p
o
r
tan
t
con
s
train
t
s are throu
ghp
u
t
, m
a
k
e
sp
an of
jo
b req
u
ests, d
e
ad
lin
e,
d
a
talo
cality cost and
QoS.
Enc
odi
ng R
u
l
e
:
Enco
di
n
g
i
s
t
h
e desi
gni
ng
chr
o
m
o
som
e
proces
s i
n
s
u
ch
a
m
a
nner t
h
a
t
t
a
sks an
d
resources a
r
e e
n
code
d in one
chrom
o
som
e
. E
ach c
h
rom
o
som
e
is represe
n
ted a
s
a
2
*
M
m
a
trix
, wh
ere,
M
is
th
e ch
ro
m
o
some len
g
t
h
.
Th
e first row
o
f
the
m
a
trix
rep
r
e
s
ents the re
ques
ted appli
cations, and second row
of
th
e m
a
trix
co
rrespon
d
s
to
the clo
u
d
whe
r
e
the application is exec
uted.
Here job re
quests
are re
gistered,
assi
gne
d
uni
qu
e i
d
, an
d t
h
e
n
c
ont
rol
l
e
r m
a
ps t
h
e rea
d
y
j
ob
r
e
que
st
t
o
avai
l
a
bl
e res
o
u
r
ces
usi
n
g G
A
. B
a
s
i
cal
l
y
j
o
b
req
u
e
sts are
d
i
v
i
d
e
d
i
n
to
m
u
ltip
le task
s.
Tasks accum
u
l
a
t
ed
in
every 5
s
ecs fo
rm
a b
a
tch
.
Len
g
t
h
o
f
ch
ro
m
o
so
m
e
d
e
p
e
nd
s on th
e leng
th
of task
s i
n
t
h
e
b
a
tc
h.
If
25 tas
k
s are
collected in a
batch t
h
en t
h
e
ch
ro
m
o
so
m
e
i
s
2*
25
.
W
e
hav
e
r
a
nd
o
m
ly
g
e
n
e
r
a
ted
300-
400
job
-
r
e
qu
ests f
r
o
m
2
0
user
s.
Leng
th
of
job
-
request, task, and se
rve
r
capa
c
ity are expres
sed in MIPS
. Nu
m
b
er of task
s in
a b
a
tch
is eq
u
a
l to
nu
mb
er
of
vi
rt
ual
m
achi
n
es. Out
o
f
1
0
serve
r
s, we
h
a
ve ra
nd
om
ly
categ
orized
the
m
in
to
3
d
i
fferen
t d
a
ta lo
catio
n
s
,
h
a
v
i
n
g
d
i
fferent d
a
ta lo
cality co
st.
Tabl
e 1. Al
l
o
c
a
t
i
on of
t
a
sks
t
o
di
ffe
re
nt
ser
v
ers
Job Request
A1 A2
A3
A4
A5
Task
s
CA1, IA1
CA2, IA2
CA3, IA3
CA4, IA4
CA5, IA5
Cloud Pr
ovider
(2
,7
)
(8
,9
)
(2
|
4
, 6
|
3
)
(1
|
9
, 4
|
2
)
(3
,4
)
Set o
f
cl
o
u
d
p
r
o
v
i
d
e
rs
h
a
v
i
ng sam
e
d
a
ta lo
cality co
st are group
ed
t
o
g
e
t
h
er, eg
. {1
,2
,3
},
{4
,5
,6
} and
{7
,8
,9
,1
0 } are clou
ds
h
a
v
i
ng
sam
e
d
a
ta locality. Alth
ou
gh
cu
rren
t
scenario
is cl
o
u
d
prov
id
ers on
ly
ch
arg
e
s
f
o
r
ou
tbou
nd
sto
r
ag
e wh
er
eas in
bo
und
st
o
r
ag
e is fr
ee. Bo
th
inb
ound
and
o
u
t
b
oun
d
i
n
cur
s
h
i
gh
carbo
n
fo
ot
p
r
i
n
t
i
n
b
o
t
h
i
n
b
o
u
n
d
a
n
d
out
bo
u
n
d
o
f
l
a
rge
am
ount
of
dat
a
.
In Ta
bl
e
1,
jo
b
req
u
est
A
1
i
s
di
vi
de
d i
n
t
o
C
P
U i
n
t
e
nsi
v
e a
nd
I/
O i
n
t
e
nsi
v
e jo
b t
a
s
k
s i
.
e,
C
A
1
and I
A
1
.
C
l
ou
d
pr
ovi
der
s
2
an
d
7 a
r
e t
h
e
opt
i
o
ns
avai
l
a
bl
e f
o
r
exec
u
t
i
on
of
t
h
e
j
o
b
req
u
est
A
1.
Si
m
i
l
a
rl
y
i
n
case o
f
jo
b
r
e
qu
est
A
4
, cl
ou
d pro
v
i
d
e
r 1
an
d 9 bo
t
h
ar
e
av
ailab
l
e
f
o
r
C
P
U
i
n
t
e
nsi
v
e whe
r
eas
cl
ou
d
pr
o
v
i
d
er
s 4
a
n
d 2 fo
r
I/O i
n
tensive
.
Charges
of
processing rate
de
pend
on the
da
ta loca
lity of t
h
e
datacenter,
bandwidth rate
, CPU
u
tilizatio
n
rate an
d
traffic o
f
jo
b
requ
ests.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
18
28
–
1
838
1
834
Though com
p
uting is dynam
i
c but at any instance of
tim
e
th
e lo
ad
balan
c
in
g
pro
b
l
em
is
fo
rm
u
l
ated
as “P
nu
m
b
er
o
f
job
s
sub
m
i
t
t
e
d
b
y
cloud
u
s
ers to
Q
n
u
m
b
e
r
o
f
pro
c
essing
u
n
its i
n
t
h
e clo
u
d
”
. Co
m
p
u
t
ab
ility
Vector (C
V)
of each
processi
ng
unit indicat
es stat
us of rat
e
of
utilization, expressed i
n
MIPS.
β
is ex
ecu
tio
n
co
st of th
e task
and
L is t
h
e d
e
lay co
st / pen
a
lty (to
b
e
paid
b
y
p
r
o
v
i
d
e
r in
case
d
e
liverab
les
fail to
meet
deadl
i
n
e
s
. C
u
r
r
ent
st
at
us
of
t
h
e
pr
ocessi
ng
uni
t
i
s
e
x
pres
s
e
d i
n
e
quat
i
o
n
(
1
)
as
fol
l
o
ws
. E
quat
i
o
n
(
2
)
sho
w
s
the
status of job request repre
s
ente
d b
y
a Job Statu
s
V
ector
(
J
SV
)
.
CV =
f (M
IPS
,
β
, L
)
(
1
)
JSV =
f (t, M
I
PS
j
,
AT,
d)
(2)
whe
r
e”t
” descr
i
bes
t
y
pe o
f
se
rvi
ce re
qui
re
d
by
t
h
e
j
o
b, sp
anni
ng
Iaa
S
, S
aaS
o
r
PaaS.
M
I
PS
j
represen
ts th
e
n
u
m
b
e
r of in
st
ru
ction
s
present in
th
e j
o
b
d
e
term
in
ed
b
y
th
e p
r
ocessor; AT in
d
i
cates th
e arriv
a
l ti
m
e
o
f
jo
b
in
th
e syste
m
an
d d
is th
e m
i
n
i
mu
m
t
i
m
e req
u
i
red
to
co
m
p
lete
th
e job
b
y
p
r
ocessor.
ζ
is th
e
co
st fu
n
c
ti
o
n
wh
ich
is minim
i
zed.
ζ
= w1
∗
β
(M
I
P
S
j
/ MIPS) +
w2
∗
L
(3
)
w
h
er
e
w1
an
d
w
2
ar
e pr
ed
ef
i
n
ed w
e
i
g
h
t
s,
dep
e
nd
ing
up
on the
user prefe
r
ence
or re
putat
ion
value
of t
h
e use
r
.
3.
1.
Proposed method usin
g Ge
netic
Algorithm
:
The stre
ngt
hs of Ge
netic Algor
ithm
technique is that it
works well when sea
r
ch s
p
ace is vast,
prem
at
ure co
n
v
er
ge
nce ca
n
be a
voi
ded
by
dy
nam
i
cal
ly
chan
gi
n
g
t
h
e
ope
rat
o
rs
of
n
a
t
u
ral
sel
ect
i
o
n.
St
eps
in
clu
d
e
s as
fo
llo
ws:
1.
Initi
a
liza
t
io
n
: Fix
e
d
leng
th
of chro
m
o
so
m
e
s is g
e
n
e
rated
for each
b
a
tch. In
itializatio
n
o
f
chrom
o
som
e
is random
ly generate
d. He
re
each job re
que
st
is conside
r
e
d
to be the set
of s
u
b-di
vide
d tasks
called
g
e
n
o
m
e. Fitn
ess
v
a
l
u
es are calculated
u
s
ing
eq
u
a
tion
(2).
Elitis
m
se
lecti
o
n
is
u
s
ed
i.e., th
e
chr
o
m
o
som
e
h
a
vi
n
g
hi
g
h
est
fi
t
n
ess val
u
e i
s
sel
ect
ed and
m
oved t
o
t
h
e next
ge
ne
rat
i
o
n m
a
t
i
ng po
ol
.
If t
h
e
b
e
st fit ch
ro
mo
so
m
e
o
f
th
e
cu
rren
t
g
e
n
e
ratio
n
fitn
ess
v
a
lu
e is lesser th
an
th
e m
o
st fit
ch
ro
m
o
so
m
e
o
f
th
e
pre
v
i
o
us ge
ne
rat
i
o
n
,
t
h
e l
a
t
e
r w
oul
d o
u
st
t
h
e chr
o
m
o
som
e
havi
ng l
east
fi
t
n
ess v
a
l
u
e of t
h
e c
u
r
r
ent
population. Some pre-re
quisit
e
data
a
r
e inde
x
of datace
nters, the
list
of VM for eac
h se
rver/
d
atacenter,
Requirem
e
nts
of each VMs
a
r
e:
TCT
tvd
=VMC
vd +
Sin
tvd +
E
tvd +
Sout
tvd
(4)
TCT
tvd
=Task c
o
m
p
l
e
t
i
on t
i
m
e
, w
h
e
n
jo
b
r
e
que
st
t
o
n
vi
rt
ual
m
achi
n
e
v
on
dat
a
ce
n
t
er d
com
p
l
e
t
e
d i
t
s
execution.
VM
C
vd
is th
e ti
me to
create
v
i
rtu
a
l m
achines in
the ph
ysical machine.
E
tvd
is th
e ex
ecu
tio
n ti
m
e
th
e task
t
run
s
on
t
h
e virt
ual m
achine
v
of datace
nter d.
Sin
tvd
is the process
of
getting all the re
quis
ite input files
a
nd Sout
is
the process of
transferi
ng t
h
e exe
c
uted
resul
t
t
o
t
h
e
p
r
edefi
n
ed
dest
i
n
at
i
ons.
R
eal
n
u
m
ber enco
di
n
g
i
s
use
d
i
n
ou
r
sim
u
l
a
t
i
on.
2.
Cr
oss
o
ver
:
The o
b
j
ect
i
v
e
of t
h
i
s
st
ep i
s
t
o
p
r
o
d
u
ce ne
w
i
ndi
vi
d
u
al
s by
refo
rm
i
ng part
s of sel
ect
ed
t
w
o ch
r
o
m
o
som
e
s. Two c
h
r
o
m
o
so
m
e
s from
t
h
e
m
a
t
i
ng po
ol
are sel
ect
ed
and e
x
cha
n
ge
d o
f
ge
nom
e part
s at
o
n
e
or m
u
ltip
le p
o
i
n
t
.
In
th
is p
a
p
e
r
on
e-po
in
t cro
s
sover is i
m
p
l
e
m
e
n
ted
wh
ic
h
is selected
ran
d
o
m
l
y
.
Prob
ab
lility o
f
cro
s
sov
e
r
(p
c
)
an
d
p
r
ob
ab
ility o
f
m
u
tatio
n
(p
m
) influences t
h
e de
gree
of s
o
lution accura
cy and
rate of co
nv
ergen
ce in im
p
l
e
m
en
tin
g
GA.
3.
Mut
a
tion
:
As we
ha
ve u
s
ed real
num
ber enc
o
di
n
g
, m
u
t
a
t
i
on i
s
di
ffe
rent
t
h
a
n
bi
nar
y
enco
di
n
g
.
Unl
i
k
e
bi
nary
bi
t
s
o
f
c
h
r
o
m
o
som
e
s are n
o
t
t
o
g
g
l
e
d
f
r
om
0/
1
or
1/
0.
Selectio
n
,
crossov
e
r and
mu
tatio
n
are
u
s
ed
iterativ
ely till th
e o
p
timal so
lu
tion
is o
b
t
ai
n
e
d
or
m
a
xim
u
m
nu
m
b
er of i
t
e
rat
i
o
n
i
s
reac
he
d.
The
fi
nal
c
h
r
o
m
o
som
e
obt
ai
ned i
s
t
h
e
be
st
po
ssi
bl
e m
a
ppi
n
g
so
lu
tion
fo
r t
h
e b
a
tch
o
f
task
s co
llected
per th
at in
stan
ce o
f
tim
e. A h
ypo
th
etical co
nfigu
r
ation
has b
een
devel
ope
d.
We have considered 3
regi
ons. Servers/
datacenters are
di
stribute
d
in these regions havi
ng
resp
ectiv
e data lo
cality co
st.
Sin
g
l
e tim
e zo
n
e
h
a
s
b
e
en
con
s
id
ered fo
r all reg
i
on
s.
Length of
eac
h
job request va
ries
betwe
e
n 5 to
20
MIPS.
Num
b
er of dat
acenters
c
o
nsidere
d
ra
nges
fr
om
2 t
o
20. Each p
h
y
s
i
cal
m
achi
n
e can p
r
o
d
u
ce VM
up
t
o
10
.
W
e
ha
v
e
rand
om
l
y
assi
gne
d dat
a
l
o
cat
i
on
cost
t
o
t
h
e
ge
o
g
ra
p
h
i
cal
l
y
di
st
ri
but
e
d
dat
a
c
e
nt
ers.
4.
Selection
:
Mo
st fit ch
ro
mo
so
m
e
is selected
an
d tak
e
n
t
o
th
e
n
e
x
t
g
e
n
e
ratio
n
m
a
tin
g
po
o
l
. Ellitis
m
selectio
n
was
u
s
ed
. Th
e abov
e
p
r
o
cesses are iterativ
el
y ex
ecu
ted
till num
b
e
r o
f
g
e
n
e
ratio
n
is attained
o
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Disaster Recov
e
ry
Services
in Intercloud Usi
n
g
Gen
e
tic Algo
rithm Loa
d
Ba
lan
cer (Ta
m
an
na
Jena
)
1
835
co
nv
erg
e
n
ce
o
f
ou
tpu
t
in
cu
r. It is tested
and
p
r
ov
ed
t
h
eo
retically th
at g
e
n
e
t
i
c alg
o
rith
m
with
ellitis
m
sele
ctio
n
ope
rat
o
r have
great
gl
o
b
al
c
o
nve
r
g
ence
.
4.
RESULTS
A
N
D
DI
SC
US
S
I
ON
Test
i
ng t
h
e p
r
o
pos
ed m
odel
o
n
real
sy
st
em
bei
ng i
n
-e
xecut
a
bl
e, si
m
u
l
a
t
i
on i
s
d
o
n
e u
s
i
n
g
M
A
TL
AB
.
Lo
ad
b
a
lan
c
ing
in
cl
o
u
d
p
l
atform
is a NP co
m
p
lete p
r
ob
lem in
real ti
m
e
.
Techn
i
qu
es
p
r
o
p
o
s
ed
till d
a
te
own
i
t
s
share of st
r
e
ngt
h an
d wea
kne
sses. O
u
r wo
rk acc
om
od
at
es opt
i
m
al
t
r
ade o
ff bet
w
ee
n:
arri
val
rat
e
of j
o
b-
r
e
qu
ests,
nu
m
b
er
of
activ
e physical
m
ach
in
es and
r
eal ti
m
e
lo
ad b
a
lan
c
ing
u
s
ing
G
A
alo
n
g
w
ith
h
i
gh
ser
v
er
u
tilizatio
n
rate
with
ou
t co
m
p
ro
m
i
sin
g
on
QoS.
First g
e
n
e
ration
in
flu
e
n
ces the q
u
a
lity o
f
th
e fin
a
l o
u
t
co
m
e
p
r
odu
ced
in
co
n
c
l
u
d
i
n
g
g
e
n
e
ration
.
It is
o
n
e
o
f
th
e v
ital step
in
th
e en
tire alg
o
r
ith
m
.
In
th
is p
a
p
e
r, in
itializatio
n
is
d
o
n
e
ran
d
o
m
ly
to
av
o
i
d
prematu
r
e
con
v
e
r
ge
nce a
n
d
G
A
o
p
erat
ors
are
sel
ect
ed
by
ri
go
r
ous
sim
u
l
a
t
i
on, i
n
or
der
t
o
o
b
t
a
i
n
opt
i
m
u
m
m
a
pp
i
n
g
com
b
i
n
at
i
on b
e
t
w
een
num
ber of act
i
v
e se
r
v
ers a
n
d im
pat
i
e
nt
t
a
sks kee
p
i
ng
hi
g
h
t
h
r
o
u
g
h
p
u
t
an
d m
i
nim
u
m
mak
e
sp
an
ti
me. Each
p
o
s
sible
m
a
p
p
i
n
g
is
assessed
u
n
d
e
r
m
u
ltip
le crite
ria. Set o
f
all
o
catio
ns ob
tain
ed
i
n
si
m
u
latio
n
sh
ows p
a
reto
efficien
t. In
acco
r
dan
ce to
Paret
o
set allo
catio
n
of reso
urces, it is i
m
p
o
ssib
l
e to
m
a
k
e
an
y op
tion
b
e
st with
ou
t m
a
k
i
n
g
atleast on
e
o
p
tion
wo
rst.
Sub
s
ets
o
f
po
ssib
le op
tion
s
are also id
en
tified
.
By
g
a
th
ering
all of th
e co
n
c
iev
e
ab
ly o
p
tim
al so
l
u
tio
ns, a
d
e
sign
er can
m
a
k
e
fo
cused
and
trialed
trad
eoffs
with
in
th
is con
s
trai
n
e
d
set
o
f
p
a
ram
e
ters,
rath
er th
an
n
eed
i
n
g to
con
s
id
er th
e fu
ll ran
g
e
s
o
f
p
a
rameters.
We f
o
u
n
d
t
h
at
our m
odel
sh
ows
bet
t
e
r l
o
a
d
- bal
a
nci
n
g r
e
sul
t
s
and e
v
e
n
su
gge
st
s hea
l
t
h
y
bal
a
nce
bet
w
ee
n
n
u
m
b
ers
of
p
h
y
s
i
cal
m
achi
n
es re
qu
i
r
ed
wi
t
h
res
p
e
c
t
t
o
ar
ri
val
rat
e
o
f
jo
b
-re
que
s
t
s. M
o
re
num
ber
of
act
i
v
e ser
v
ers
doe
s n
o
t
c
ont
r
i
but
e a
n
y
val
u
e ad
di
t
i
on
o
n
l
y
enha
nces t
h
e m
a
i
n
t
e
nance
cost
.
He
re
w
e
ha
ve
t
a
ken j
o
b
-
req
u
e
st
s
o
f
di
ver
s
e l
e
ngt
h
.
We fo
u
n
d
t
h
at
s
w
i
t
c
h on/
of
f of
p
h
y
s
i
cal
m
achi
n
e
i
s
a
be
nefi
ci
al
p
r
op
osa
l
o
n
l
y wh
en
it can
rem
a
in
s id
l
e
fo
r a lon
g
time( en
ou
gh
to
co
m
p
en
siate th
e switch
on
/o
ff o
v
e
rh
ead
)
an
d it a
l
so
suf
f
ers
l
o
ng
w
a
i
t
i
ng o
f
rea
d
y
j
obs i
n
que
ue
by
t
h
e t
i
m
e phy
si
cal
m
achi
n
e i
s
l
i
n
ed
up
f
o
r se
r
v
i
ce. [
3
1
]
have
co
m
p
ared
ad
ap
tiv
e
g
e
n
e
tic alg
o
rith
m
(AGA) and
ad
ap
ti
v
e
algo
r
i
t
h
m
-
j
o
b sp
ann
i
ng
time an
d
l
o
ad balan
c
ing
genet
i
c
al
g
o
ri
t
h
m
(JLGA
)
t
o
achi
e
ve l
o
ad
bal
a
nci
n
g
w
ith least
m
a
k
e
sp
an
bu
t
h
a
v
e
consid
ered
o
n
l
y 30
job
-
requests using com
p
aratively
higher
n
u
m
b
er of
n
odes
.
C
o
m
p
arat
i
v
el
y
fewer
n
u
m
b
er of
jo
b
-re
que
st
s (
r
an
ges
f
r
o
m
1
0
to
100)
w
ith
less
d
i
v
e
r
s
ity in
j
ob-
r
e
qu
ests size ar
e
used
fo
r
m
a
p
p
i
ng
[2
6
]
. Main
ly
all j
o
b-
r
e
qu
est
s
ar
e
C
P
U i
n
t
e
nsi
v
e
and b
u
l
k
y
.
We ha
ve d
o
n
e
sim
u
l
a
t
i
on on
m
o
r
e
th
an
300
jo
b-
r
e
qu
ests. Jo
b-
r
e
qu
ests ar
e of
v
a
ri
o
u
s
sizes an
d cap
acity.
We
h
a
v
e
sim
u
l
a
ted
with wi
d
e
rang
e
of arrival rate, co
m
p
u
t
ab
ility o
f
serv
ers an
d
d
i
v
e
rsity in
j
o
b
sizes.
Our
m
o
d
e
l is q
u
ite flex
i
b
le and
ad
ap
ted
well
with
sud
d
e
n
ch
ang
e
of
work
lo
ad
s.
Num
b
er of job-request produced ra
nges from
200-400 in each sim
u
lation
which is furthe
r divi
de
d into
num
erous tas
k
s.
Sim
u
l
a
t
i
on res
u
l
t
sho
w
n i
n
F
i
gu
re 3 sh
ow
s,
t
h
e
m
a
kespan
t
i
m
e
of t
o
t
a
l
jo
bs (
h
ere m
a
i
n
ly
hi
gh C
P
U
i
n
t
e
nsi
v
e
j
o
b
-
r
e
que
st
s are c
o
nsi
d
e
r
ed
) w
h
e
r
e w
o
r
k
l
o
a
d
i
s
di
st
ri
b
u
t
e
d i
n
a bal
a
nce
d
w
a
y
am
ong a
v
a
i
l
a
bl
e
m
u
l
tip
le
resou
r
ces
wh
en
rate o
f
arriv
a
l o
f
j
o
b
-
requ
ests
is 10
.
X-ax
is refers to
t
h
e
num
b
er
of active se
rvers a
nd
y-ax
is expresses to
tal m
a
k
e
s
p
an
tim
e o
f
job
-
requ
ests
h
e
l
d
i
n
t
h
e bat
c
h,
whi
c
h are
m
a
ppe
d t
o
res
o
ur
ses i
n
a
bal
a
nce
d
m
a
nner u
s
i
n
g G
A
and FI
FO
. I
n
Fi
gur
e 4, w
h
ere arri
val
rat
e
i
s
20. R
e
sul
t
s sug
g
est
s
t
h
at
t
h
e
min
i
m
u
m
m
a
k
e
sp
an
tim
e can
b
e
ach
iev
e
d wh
en nu
m
b
er
o
f
activ
e serv
ers i
s
14
. Ou
r m
o
del h
e
lp
s to attain
th
e
opt
i
m
u
m
t
r
ade
off
bet
w
ee
n sup
p
l
y
and
dem
a
nd
whi
l
e
kee
p
i
ng Q
o
S i
n
t
act
and c
o
m
p
l
e
t
i
on of
jo
b ex
ecut
i
on i
n
real tim
e
. In F
i
gu
re 5
,
w
h
er
e
α
i
s
30 s
h
o
w
s
t
h
at
m
i
nim
u
m
m
a
kespa
n
t
i
m
e
of
j
ob e
x
ec
ut
i
on ca
n b
e
ach
i
e
ve
d
usi
n
g
1
2
se
rve
r
s.
Di
st
ri
b
u
t
i
o
n
of
j
o
b-
req
u
e
s
t
s
are e
ffi
ci
e
n
t
l
y
m
a
pped t
o
reso
u
r
ces.
Fi
g
u
re
6
s
h
o
w
s
t
h
e
pl
o
t
pat
t
e
rn
whe
n
a
rri
val
rat
e
i
s
4
0
, he
re o
p
t
i
m
u
m
di
st
ri
but
i
on
can be achi
e
ve
d w
h
en se
rve
r
s
used are
10
. Gra
p
h
indicates consi
d
eri
ng
10 servers is the
opti
m
u
m
solu
t
i
on
whe
r
eas c
o
n
s
i
d
eri
ng
1
6
ser
v
ers w
h
e
n
m
a
kespa
n
t
i
m
e
i
s
l
east
.
Ove
r
head
o
f
s
w
i
t
c
h
on/
of
f c
a
n
be est
i
m
at
ed
here
by
usi
n
g si
m
u
l
a
t
i
on re
sul
t
.
O
u
r m
ode
l
evenl
y
d
i
stribu
tes th
e work
l
o
ad
with
in
real ti
m
e
in
s
u
ch
a m
a
n
n
e
r th
at u
tilizatio
n
o
f
pro
cesso
rs are k
e
p
t
h
i
g
h
. It h
e
lp
s
i
n
deci
di
n
g
t
h
e pe
rf
orm
a
nce
o
f
t
h
e e
n
t
i
r
e
sy
st
em
. It
su
g
g
est
s
sm
art
de
ci
si
on m
a
ki
ng
re
gar
d
i
n
g a
v
a
i
l
a
bl
e
resources and
its u
s
ag
e.
Dep
e
n
d
i
n
g
o
n
t
h
e dyn
amic p
a
ramete
rs lik
e av
ailab
ility o
f
serv
ers, arriv
a
l rate
o
f
j
o
b
-
requ
ests and
ev
en co
m
p
u
t
ab
ility o
f
serv
ers
wh
ich
k
e
ep
o
n
ch
ang
i
ng
in real life, m
a
in
l
y
in
u
n
c
on
v
e
ntio
nal
circum
stances then the optimal scen
ari
o
can be p
r
o
act
i
v
el
y
exp
r
esse
d
usi
ng
ou
r m
odel
.
Du
ri
n
g
di
sast
er
reco
very
s
u
d
d
e
n cha
n
ge of
arri
val
rat
e
of
jo
b-
re
quest
s
can p
u
t
ext
r
a
st
ress o
n
ava
i
l
a
bl
e reso
urce
s an
d
co
m
p
letio
n
of i
m
p
o
r
tan
t
job
-req
u
e
sts with
i
n
esti
m
a
ted
d
eadlin
e is ch
allengin
g
as
well as cru
c
ial.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
18
28
–
1
838
1
836
Figure
3.
w
h
e
n
α
is
10
Fi
gure
4.
w
h
e
n
α
is 20
Figure
5.
w
h
en
α
is
30
Figure
6.
whe
n
α
is
40
In all cases
GA loa
d
balanc
er is m
o
re efficient
th
an FIFO.
Arriv
a
l
rate rang
es fro
m
4
to 40
and
d
i
v
e
rsity o
f
j
o
b
-
requ
ests ex
ists, we tried
t
o
lo
ad
b
a
lan
c
e th
e in
co
m
i
n
g
job
-
requ
ests u
s
ing
m
u
ltip
l
e
b
u
t
dy
nam
i
c num
ber
of se
rve
r
s.
W
e
t
r
i
e
d t
o
est
a
bl
i
s
h a t
r
a
d
e-
of
f bet
w
ee
n ar
ri
val
rat
e
and
n
u
m
b
er o
f
act
i
v
e
physical m
achines. Som
e
times in disa
ster
recovery s
w
itch on and s
w
itc
h off
physica
l
m
achines are less
rewa
rdi
n
g
.
P
o
wer m
a
nagem
e
nt
t
ech
ni
q
u
es
that feat feat
ures
of virt
ual
i
zation to sa
ve energy are
highly
app
r
op
ri
at
e wi
t
h
t
h
e
i
m
pat
i
e
nt scena
r
i
o
.
S
p
ec
i
a
l
foc
u
s i
s
gi
v
e
n t
o
p
o
w
er m
a
nagem
e
nt
t
echni
que
s t
h
at
e
x
pl
oi
t
t
h
e vi
rt
ual
i
zat
i
on t
e
c
h
nol
ogy
t
o
save
ene
r
gy
. A si
m
i
l
a
r wo
rk i
s
d
one i
n
[
32]
whe
r
e
hy
b
r
i
d
genet
i
c
al
g
o
ri
t
h
m
al
on
g wi
t
h
g
r
e
e
dy
al
go
ri
t
h
m
is used f
o
r res
o
urce sc
hed
u
l
i
n
g. Li
ve m
i
grat
i
on
of w
o
rkl
o
ad
from
heavi
l
y
loade
d
v
i
rtu
a
l m
ach
in
es are sh
i
f
ted to
id
l
e
o
n
es.
Thei
r
pr
op
os
ed m
odel
sho
w
s better elas
ticity but associate
d
o
v
e
rh
ead
an
d
ex
p
e
n
s
es
o
f
li
v
e
m
i
g
r
atio
n
as well as d
a
ta lo
cality co
st
wh
ich
is po
ssi
b
ility d
u
e
to
laten
c
y
arb
itrag
e is n
o
t co
n
s
id
ered. Streng
th
of ou
r
m
o
d
e
l is
n
e
g
l
i
g
ib
le ti
m
e
sp
en
t for op
tim
u
m
reso
urce m
a
p
p
i
n
g
.
Basically
it i
m
p
r
o
v
i
ses th
e resou
r
ce allocatio
n
pro
c
edure
in fast and
efficient way.
Rate of em
ptiness of
que
ue
i
s
hi
g
h
(m
appi
ng
t
ech
ni
q
u
e i
n
GA
i
s
bet
t
e
r
t
h
a
n
FIF
O
,
t
h
ere
f
or
e sam
e
bat
c
h
of
j
o
b-
req
u
est
t
a
ke
s
com
p
aratively lesser tim
e when
GA bas
e
d load bala
ncer is
use
d
) in ca
se
of GA than FIFO.
5.
CO
NCL
USI
O
N
Loa
d
bal
a
nce
r
usi
n
g G
A
pl
ay
s cruci
a
l
rol
e
i
n
per
f
o
rm
ance of i
n
t
e
rcl
o
u
d
pl
at
fo
rm
. C
o
m
b
at
pr
oces
s
of
disaster
ne
ed accurate a
nd t
h
ou
ghtful
precision in
perform
a
nce, as
incom
i
ng job
requests c
oul
d be
i
m
p
a
tien
t
an
d
erratic,
resources will b
e
li
mited
an
d
tim
e con
s
train
t
will b
e
steep
.
Estab
lish
i
ng
trad
e
o
f
f
bet
w
ee
n
dem
a
nd
w
h
i
c
h i
n
di
cat
e im
pat
i
e
nt
job
-re
q
u
est
s
a
n
d
s
u
pply speci
fies available
r
e
so
ur
ces h
e
r
e
is to
ug
h,
whe
n
rat
e
o
f
ar
ri
val
of
j
obs i
s
spo
r
a
d
i
c
, dea
d
l
i
nes are st
er
n and ca
rb
o
n
-
f
o
o
t
-
p
r
i
n
t
i
s
t
o
be r
e
st
ri
ct
ed. G
A
b
a
sed
lo
ad
b
a
lan
c
er sh
ows
b
e
tter
resu
lt th
an trad
itio
n
a
lly u
s
e
d
F
I
FO rega
r
d
i
n
g opt
i
m
u
m
al
l
o
cat
i
on of j
o
b
-re
que
st
s
to resources. Queui
ng
use
d
he
re helps in effi
cient
task
m
a
pping am
ong m
u
ltiple hom
oge
neous serve
r
s havi
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Disaster Recov
e
ry
Services
in Intercloud Usi
n
g
Gen
e
tic Algo
rithm Loa
d
Ba
lan
cer (Ta
m
an
na
Jena
)
1
837
d
i
fferen
t
d
a
ta lo
cality co
st a
n
d
strict real ti
m
e
c
o
n
s
train
t
. GA b
a
sed
map
p
i
n
g
p
r
o
c
edu
r
e is v
e
ry fast an
d
efficiently
distrib
u
te wo
rkl
o
a
d
s ev
e
n
l
y
am
on
g p
r
ocess
o
r
s
. Deci
si
o
n
va
riables considered are CPU usage
,
th
ro
ugh
pu
t, d
a
ta lo
cality
co
st an
d
m
a
k
e
sp
an
ti
m
e
. Th
roug
hpu
t is h
i
g
h
e
r in
GA lo
ad
b
a
lan
c
er th
an
FIFO.
Serv
er
u
tilizati
o
n rate is
fo
und
to b
e
h
i
g
h
in all cases
wh
en
GA b
a
sed
l
o
ad
b
a
lan
c
er is
u
s
ed
resu
lting
lo
wer
car
bon
foo
t
-p
r
i
n
t
.
REFERE
NC
ES
[1]
Orgerie A. C.,
et al.
, “A survey
on
techniques
for improving
the en
er
g
y
effic
i
enc
y
of
la
rge-s
cal
e dis
t
rib
u
ted
s
y
ste
m
s,
”
ACM
Computing Surveys (
C
SUR)
,
vol/issue: 46(4), pp.
47, 2014
.
[2]
Ma
ste
l
ic
T.,
et a
l
.
, “Cloud Computing: Survey
on
Energ
y
Eff
i
cien
cy
,”
ACM Computing Sur
veys
(
C
SUR)
,
vol/issue:
47(2), pp
. 33
, 20
15.
[3]
“
A
ccenture
in
coll
aborat
ion with W
S
P
Environm
ent & En
erg
y
,”
C
l
oud
Computing Sustainability: The
Environmenta
l
Bene
fits of
Mov
i
ng to
the Clou
d,
2010. http
://
nstore.a
cc
enture
.com
/.../c
cr/2010
-2011/Accen
ture
-
Sustainability
-Cloud-Computing-TheEnv
ir
onmentalBen
efitsofMov
i
ngtotheC
l
oud.p
d
f.
[4]
Bernstein D.,
et al.
, “Bluepr
i
nt f
o
r the inter
c
loud
-protocols and
f
o
rmats for cloud computi
ng inter
operability
,
” in
Internet and W
e
b Applications
a
nd Services. ICI
W
'09. Fourth
International Conference I
E
EE
, pp.
328-336, 2009
.
[5]
Abawajy
J., “Determining servi
ce trustworthin
ess in interc
loud computing envir
onments,”
in
Perv
asiv
e
Sy
ste
m
s,
Algorithms, and
Networks (
I
SPAN)
,
10th International S
y
mposium IEEE
, pp
. 784
-788, 2009
.
[6]
Bitten
c
ourt L
.
F., “
M
adeira ER
, Da F
onseca NL. Scheduling in h
y
brid
clouds,”
C
o
mmunications Magazine, IEEE
,
vol/issue: 50(9), pp.
42-47
,
2012
.
[7]
Toosi A. N., “Calheiros RN, Buyy
a
R
.
Interconn
ected cloud
computing envi
ronm
ents: Challeng
es, taxonom
y
,
an
d
survey
,
”
ACM C
o
mputing Surveys (
C
SUR
)
, vol/is
sue: 47(1), pp
. 7
,
2014.
[8]
J
.
F
.
M
o
r
a
i
s
a
n
d
G
.
L
.
S
i
queira
,
“
W
i
reles
s
techn
o
logies
environ
m
ental im
pacts
,
”
Micr
owave an
d Optoelec
tr
oni
cs
Confer
enc
e
(
IMOC)
SBMO/IEEE MTT-S Intern
ational
, pp
. 523-
527, 2009
.
[9]
Brooks T.,
et a
l
.
, “Conceptualizing a secure
wireless cloud
,”
Internationa
l Journal of Clou
d Computing and
Services Science (
I
J-CLOSER)
, vol/issue: 1
(
3), pp
. 89-114
, 2012
.
[10]
Chauhan S., “Environmental
and Health Hazards of Mobile Devi
ces and
Wireless Communication,” 200
2.
http://cr
y
stal.u
ta.edu/~kumar
/cse6392/.../Sav
ita_p
aper.pdf
.
[11]
Berm
údez A,
et al.
, “Fast routin
g computation
o
n
InfiniBand networks,”
Parallel and Distributed
Systems, IEEE
Transactions,
vo
l/issue: 17
(3), pp
. 215-226
, 2006
.
[12]
Demchenko Y.,
et al.
, “Defining Inter-Cloud
architecture fo
r
Interoper
a
bility and Integration
,
”
The 4th IEEE
Conference on
C
l
oud Computing
,
(
C
loudCom201
2)
, Taip
ei, Taiw
an, 2012
.
[13]
Rathor V. S.,
et al.
, “Survey
on Load Balan
c
ing through V
i
rtual M
achin
e
Scheduling in
cloud computing
Environment,”
I
n
ternational Jou
r
nal of C
l
oud
Co
mputing and S
e
rvices Science,
vo
l/issue: 3
(
1), pp.
37, 2014
.
[14]
De Falco I,
et al.
, “Two new
fast heur
istics
f
o
r mapping par
a
llel
appli
cation
s
on cloud
computing,”
Fu
ture
Generation Com
puter Systems 37
, vol. 37
, pp
. 1-1
3
, 2014
.
[15]
S.
Va
rma
P.
,
et al.
, “Performance analy
s
is of cloud co
mputing using queuing models,” in
Cloud Computing
Technologies, Applic
ations and
Management (
I
CCCTAM)
,
2012
I
n
ternational Co
nference I
EEE,
pp. 12-15
, 2012
.
[16]
Chiang Y. J
.,
et al.
, “Performance
and Cost-eff
ectiv
eness Analy
s
es for C
l
oud
Services B
a
sed
on Rejected an
d
Impatient Users,”
SeviceComputing,
IEEE Transactions on
Serv
ic
es Computing
, v
o
l. 99
, 2014
.
[17]
Mehdi N.,
et al.
, “Virtual machines cooperation for
impatient jobs
under cloud Paradigm,”
International Journal of
Information
and Communication Engineering,
vol/issue: 7(1)
, pp
.
13-9, 2011
.
[18]
S
e
lvaran
i S
.
and
S
a
dhas
i
vam
G.
S
., “
I
m
p
roved
cos
t
-bas
ed
algor
ithm for task scheduling
in clou
d computing,” in
Computational i
n
tell
igen
ce
and computing
resea
r
ch (
I
CCIC)
, IEEE
internat
ional
conf
erenc
e
on
,
pp. 1-5
,
2010
.
[19]
Yamini R., “Po
w
er management in
cloud computing using green algorithm,”
In
Advances
in En
gineer
ing
,
Scien
ce
and Managemen
t
(
I
CAESM)
, IEEE International
Conference on
,
pp. 128-133
, 20
12.
[20]
Sotiriadis S
.
,
et al.
, “Towards in
ter-cloud simulation p
e
rformance an
al
y
s
is: Exploring service-or
iented
ben
c
hmarks
of clouds in Si
mIC,”
InAdvanc
ed Information Networking and
Applicat
ions Workshops
(
W
A
I
NA)
,
IEEE 27t
h
International Co
nference on
,
pp
. 765-771,
2013
.
[21]
J
u
s
t
afort V. D. and P
i
erre S
., “
P
erform
ance-aware vi
rtu
a
l
m
achine al
loca
t
i
on approach i
n
an interc
loud
environment,” in
Electrica
l
&
Co
mputer Engineering (
CCECE)
,
2012 25th IEEE
Canadian Conference on,
pp
. 1-4
,
2012.
[22]
Mc
Gra
t
h M.
G.
,
et al.
, “Interclo
ud Networks
Performance Analy
s
is,”
InCloud Engine
ering (
I
C
2
E)
, 2015 IEEE
International Co
nference on
, pp
.
487-492, 2015
.
[23]
Al Sallami N.
and Al Aloussi S., “A ge
netic algorithm in green cloud computing,”
British Jo
urnal of Appli
e
d
Scien
c
e &
Technology
, vol/issue:
7(2), pp
. 179
, 20
15.
[24]
Kumar V.
V.
and Dinesh K.
,
“Job sc
heduling
using fuzzy
neural n
e
twor
k algorithm in clou
d environment,”
Bonfring In
terna
tional
Journal
o
f
Man Ma
chine I
n
terface
, vol/issue: 2(1)
, pp
. 1-6
,
2012.
[25]
Mehdi N. A.,
et al.
, “Impatien
t
task mapping in elas
tic cloud
using genetic algorithm,”
Journ
a
l of Compute
r
Scien
c
e
, vol/issue: 7(6)
, pp
. 877
,
2011.
[26]
Ra
na
M.
M.
,
et al.
, “A stud
y
on load balan
c
ing
in cloud computi
ng environment using evolution
a
r
y
and swarm
based algorithms,” in
Control, Instrumentation,
Communication
and Computa
tional Technolo
g
ies (
I
CCICCT),
IEEE Internatio
nal
Conf
er
ence
on,
pp
. 245-250
, 2014.
Evaluation Warning : The document was created with Spire.PDF for Python.