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.
8
, No
.
6
,
Decem
ber
201
8
, p
p.
4646
~
4653
IS
S
N:
20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v
8
i
6
.
pp
4646
-
46
53
4646
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
A Comparison
of Cloud E
xecution
Mechan
isms Fog
, Edge,
and
Clone Cl
oud C
om
pu
ting
T. Fra
ncis
Dep
a
rtm
ent o
f C
om
pu
te
r
E
ng
i
neer
i
ng,
BITS
Pila
ni
Uni
ver
sit
y
,
Indi
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
r 27
, 201
8
Re
vised
Jun
7
,
201
8
Accepte
d
Se
p 9
, 2
01
8
Cloud
computing
is
a
te
chnol
og
y
that
was
deve
l
oped
a
dec
ad
e
a
go
to
provide
unint
err
up
te
d,
sc
al
ab
le
se
rvices
t
o
users
and
orga
niz
a
ti
ons.
C
loud
computing
has
al
so
b
ec
om
e
an
a
t
tracti
v
e
f
ea
tur
e
for
m
obil
e
users
du
e
to
the
li
m
it
e
d
fea
tur
es
of
m
obil
e
dev
ices.
Th
e
combination
of
cl
oud
techno
logi
es
with
m
obil
e
te
chnol
o
gie
s
result
ed
in
a
new
are
a
of
computing
ca
ll
ed
m
obil
e
cl
ou
d
computing.
Th
i
s
combined
te
c
hnolog
y
is
used
to
augment
t
h
e
resourc
e
s
exi
sting
in
Sm
art
devi
c
es.
In
recent
ti
m
es,
Fog
c
om
puti
ng,
Edge
computing,
and
Clone
Clou
d
computing
t
echnique
s
have
be
come
the
la
t
est
tre
nds
after
m
obil
e
c
loud
c
om
puti
ng,
whic
h
have
a
ll
b
een
develope
d
to
addr
ess
th
e
li
m
it
ations
in
cloud
compu
ti
ng.
Thi
s
paper
rev
iew
s
the
se
re
ce
nt
t
ec
hnolog
ie
s
in
det
a
il
and
pr
ovide
s
a
compa
rat
iv
e
stud
y
of
the
m
.
It
al
so
ad
dre
ss
es
the
diffe
ren
ce
s
in
t
hese
te
chno
logies
and
how
ea
ch
of
the
m
is
eff
ective
for
orga
nizati
ons
an
d
develope
rs.
Ke
yw
or
d:
Cl
on
e cl
oud
Ed
ge
c
om
pu
ti
ng
Ela
sti
c execu
ti
on
s
Fog c
om
pu
ti
ng
IoT
Mob
il
e
back
-
e
nd
s
Mob
il
e clo
ud c
om
pu
ti
ng
Mob
il
e
off
-
loa
ding m
echan
is
m
s
Virtuali
zat
ion
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
T. Francis
,
Dep
a
rt
m
ent o
f El
ect
rical
an
d
Com
pu
te
r
E
ng
i
neer
i
ng
,
Dep
a
rtm
ent o
f C
om
pu
te
r
E
ng
i
neer
i
ng,
BITS P
il
a
ni Unive
rsity
,
I
nd
ia
.
Em
a
il
:
t
inajibu
@g
m
ai
l.co
m
1.
INTROD
U
CTION
Sm
art
dev
ic
es
su
c
h
as,
Sm
art
phones,
ta
blets,
PCs,
Net
-
books,
et
c.
,
are
tre
nd
i
ng
i
n
the
m
ark
et
to
day.
These
dev
ic
es
are
com
pact
and
sm
al
l,
whic
h
m
ake
them
han
dy
f
or
us
e.
Be
in
g
sm
al
l,
they
hav
e
few
e
r
resou
rces
than
oth
e
r
syst
e
m
s.
Cl
oud
com
pu
ti
ng
was
us
e
d
to
add
re
ss
the
dr
awb
ac
ks
in
the
se
dev
ic
es,
s
uc
h
as
la
ck
of
sto
rage,
com
pu
ta
ti
onal
powe
r,
a
nd
the
li
m
it
ed
batte
ry
li
fe
of
Sm
art
dev
i
ces[1
]
.
M
ob
il
e
cl
oud
com
pu
ti
ng
is
an
inter
discipli
nar
y
com
bin
at
ion
of
cl
ou
d
and
m
ob
il
e
com
pu
ti
ng
.
Re
s
ource
co
ns
trai
nts
in
m
ob
il
es
gav
e
rise
to
the
us
e
of
the
cl
oud
in
m
ob
il
es,
wh
ic
h
was
la
te
r
known
as
Mob
il
e
Cl
ou
d
com
pu
t
ing
or
MC
C
[2
]
.
It
is
al
so
c
on
si
der
e
d
as
an
e
xtensi
on
to
cl
oud
co
m
pu
ti
ng
,
but
a
t
the
sam
e
tim
e,
it
pro
vi
des
s
erv
ic
e
s
to
Sm
art
dev
ic
es.
The
re
are
a
nu
m
ber
of
li
m
it
a
ti
on
s
in
usi
ng
m
ob
il
e
cl
ou
d
c
om
pu
ti
ng
,
as
cl
ou
d
data
center
s
are
of
te
n l
ocat
ed
at
rem
ote locati
on
s
from
the p
la
ce
of r
e
qu
est
.
This
m
akes
it
m
or
e
su
scepti
ble
to
suffe
r
f
r
om
la
g,
network
resil
ie
nce,
a
nd
li
nk
fail
ure
s,
thus,
the
obj
ect
ive
of
usi
ng
m
ob
il
e
cl
o
ud
c
om
pu
ti
ng
fail
s
beca
us
e
of
the
se
iss
ues.
Also
the
volu
m
e
and
s
peed
of
data
in
ap
plica
ti
on
s
su
c
h
as
healt
hcar
e
,
m
anu
fa
ct
ur
in
g
a
nd
t
ra
ff
ic
m
anag
em
e
nt
is
high
hence
big
data
an
al
yt
ic
s
too
ls
su
c
h
as
m
aho
ut,
s
pa
rk
s
et
c
are
us
ed.T
hey
process
m
ulti
ple
data
sou
rces
,
distri
bu
t
ed
m
e
m
or
y
ta
s
ks
an
d
range
of
al
gorith
m
s
ho
wever
it
is
st
il
l
i
nclined
to
cl
oud
data
centr
es
who
pro
vi
de
an
d
sup
port
their
serv
ic
es
[
3].
H
ence
the
re
is
ne
ed
to
get
t
he
cl
oud
cl
ose
r
t
o
us
e
rs.
Ed
ge,
F
og,
a
nd
Cl
one
cl
ouds
try
t
o
a
ddress
these
iss
ues
an
d
are
dif
fe
ren
ti
at
ed
from
each
oth
er
in
te
rm
s
of
us
a
ge
a
nd
bu
il
d
te
c
hn
i
ques.
The
m
os
t
recent
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
A Co
mparis
on
of Clo
ud Exec
ution Mec
hani
sms
F
og, E
dge
, and
…
(
T.
Fr
an
ci
s
)
4647
dev
el
op
m
ents
in
cl
oud
c
om
puti
ng
are
fog
a
nd
ed
ge
c
om
pu
ti
ng
.
It
is
neces
sary
to
m
ake
a
com
par
ison
of
these
te
chn
iq
ues
to u
nd
e
rstan
d
t
he
a
dv
a
ntage
s
of
usi
ng ed
ge, f
og, a
nd clo
ne
cl
oud
in
d
i
ff
e
ren
t s
yst
e
m
s.
Cl
oud
-
base
d
app
li
cat
io
ns
us
e
a
central
server
to
process
da
ta
,
wh
ic
h
i
ncrea
ses
the
com
m
un
ic
at
ion
betwee
n
us
e
r
de
vices.
He
nce,
there
is
a
nee
d
for
looki
ng
b
e
yond
the
cl
oud
at
the
edg
e
of cloud
net
works.
Th
e
aim
is
to
exp
l
or
e
possibil
it
ie
s
of
pe
rfor
m
ing
com
pu
ta
ti
on
s
at
edg
e
no
de
s.
Ed
ge
Cl
ou
d
aug
m
ents
trad
it
ion
al
data
centers
c
onsist
ing
of
cl
oud
m
od
el
s,
with
ser
vice
node
s
placed
at
the
netw
ork
ed
ges
[4
]
[
5].
The
pro
xim
it
y
of
e
dg
e
nodes,
al
lows
data
processin
g
to
an
d
from
re
m
ote
cl
ouds
to
be
done
at
the
ed
ge
.
Com
pu
ti
ng
on
ed
ge
nodes
close
r
to
end
-
us
e
rs
ca
n be e
xp
l
oited as
a p
la
tf
or
m
f
or
app
li
cat
io
n pro
vid
e
rs
to
im
pr
ove thei
r
se
rv
ic
e
s.
Si
m
il
arly
,
cl
on
e
cl
ouds
pro
vi
de
a
distrib
ute
d
m
echan
is
m
of
a
pp
li
cat
io
n
execu
ti
on,
i
n
con
t
rast
to
edg
e
cl
oud.
It
autom
at
ic
ally
t
ran
s
f
or
m
s
m
ob
il
e
app
li
cat
ion
s
to
be
ne
fit
fro
m
the
cl
oud
[
6]
.
This
a
pp
li
cat
ion
i
s
a
flexi
ble
r
un
-
t
i
m
e
execu
ti
on
par
ti
ti
on
e
r,
w
hi
ch
tra
ns
fe
rs
e
xecu
ti
on
onto
dev
ic
e
cl
on
e
s
r
unning
on
the
cl
oud.
A
cl
one
al
lo
w
s
for
t
h
e
dy
na
m
ic
execu
ti
on
of
va
rio
us
a
ppli
cat
ion
s
by
al
te
rn
at
in
g
betwe
en
the
cl
one
a
nd
the
dev
ic
e.
In
the
sam
e
way,
f
og
c
ompu
ti
ng
giv
es
t
he
us
e
r
the
opti
on
of
perf
orm
ing
cl
ou
d
operati
ons
at
locat
ion
s
cl
os
e
r
to
his
or
her
point
of
inter
est
.
F
og
com
pu
ti
ng
is
sim
i
lar
in
m
any
ways
to
cl
oud
a
nd
ed
ge
com
pu
ti
ng
.
T
he
hu
ge
in
flu
x
of
real
-
ti
m
e
data,
a
nd
the
need
f
or
proc
essing
the
sa
m
e,
ga
ve
rise
to
t
he
te
rm
ino
lo
gy
of
f
og
com
pu
ti
ng
[7
]
.
F
og
no
des
a
re
heter
ogene
ous
de
vices,
r
an
ging
fro
m
hig
h
-
e
nd
se
rv
e
rs
,
acce
ss
points,
set
-
top
boxes
,
edg
e
router
s,
e
tc
.,
to
en
d
de
vi
ces,
su
c
h
as
m
ob
il
e
ph
ones
,
sm
art
watches,
an
d
sens
or
s
. I
t
us
es
ex
ist
ing
netw
orks
a
nd ro
uters
in
nea
rb
y l
oca
ti
on
s to per
f
orm
o
per
at
io
ns
just li
ke
the clo
ud
[
8].
Fog
c
om
pu
ti
ng
al
so
has
a
bett
er
res
ponse
to
the
I
nte
r
net
of
Things
e
nvir
onm
ent
and
is
s
ui
ta
ble
for
real
-
t
i
m
e
serv
ic
e
re
qu
est
s.
The
goal
of
thi
s
stu
dy
is
to
unde
rstan
d
t
he
diff
e
re
nces
bet
ween
these
te
chnolo
gies
by
pe
rfor
m
ing
a
com
par
at
ive
stud
y
of
each
of
these
appro
ac
hes.
S
om
e
of
t
he
ap
plica
ti
on
areas
w
her
e
th
ey
are
us
efu
l
a
re
al
so
discusse
d.The
nex
t
sect
ion
pro
vid
es
a
s
urv
ey
on
e
dg
e
,
fog,
a
nd
t
he
c
om
po
nen
ts
of
t
heir
buil
ds
t
ha
t
m
ake
them
feasible
for
im
ple
m
ent
at
ion
.
T
his
st
udy
inclu
des
a
com
par
at
ive
stud
y
on
ed
ge
com
pu
ti
ng
a
nd
f
og
com
pu
ti
ng
,
a di
scussion
o
n
th
e
el
ast
ic
execu
ti
on
m
echan
ism
s
us
in
g
cl
one
c
loud
com
pu
ti
ng,
a
nd
a
n
ins
pe
ct
io
n
of
off
-
loa
ding
m
echan
ism
s
us
ed
in
cl
ones
a
nd
the
co
st
-
be
ne
fit
analy
sis
to
off
-
l
oa
d.
A
dd
i
ti
on
al
ly
,
a
s
umm
ary
and com
par
at
ive st
ud
y
on fo
g,
e
dge, a
nd cl
on
e
clo
ud m
ec
han
ism
s is
pro
vid
e
d.
2.
SURVE
Y
O
N E
DGE
COMP
UTING
AND F
OG
COMP
UTING
2.1.
Edge
Compu
t
ing
Rev
ie
w
It
is
kn
own
t
ha
t
cl
oud
c
om
pu
ti
ng
an
d
m
obil
e
com
pu
ti
ng
are
us
ed
to
get
her
to
harness
m
ob
il
e
back
-
ends
to
au
gm
e
nt
the
res
ource
s
for
Sm
art
dev
ic
es.
H
oweve
r,
they
suffe
r
f
r
om
la
g,
netw
ork
resil
ie
nce,
a
nd
li
nk
fail
ur
es
.
E
dg
e
com
pu
ti
ng
is
a
te
rm
ino
lo
gy
that
is
us
e
d
t
o
a
ug
m
ent
tradit
ion
al
data
ce
ntr
es
with
se
r
vice
node
s
at
network
e
dges.
Mob
il
e
Ed
ge
Com
pu
ti
ng
(MEC)
[
9]
al
l
ow
s
t
he
us
e
of
cl
ou
d
a
nd
IT
serv
ic
es
within
cl
ose
pro
xim
it
y
of
m
ob
il
e
su
bs
cri
ber
s
,
th
us
re
du
ci
ng
the
e
nd
-
to
-
e
nd
la
te
ncy.
It
is
base
d
on
a
decen
t
rali
zed
m
od
el
that i
nterc
onne
ct
s a h
et
er
oge
ne
ous clo
ud a
nd it
is b
ase
d on the
fo
ll
owin
g
el
e
m
ents:
a.
Pr
oxim
ity:
Th
e possibil
it
y t
o use
nodes
that
are
near
e
r
r
at
he
r
tha
n fart
her
away.
b.
In
te
ll
igence:
U
se of a
utono
m
ou
s
dec
isi
on m
akin
g
to
h
el
p
i
n
the
m
iniat
ur
iz
at
ion
of syst
em
s.
c.
Con
tr
ol:
Ma
na
gem
ent
and
co
ordinati
on
com
es
fr
om
ed
ge
m
achines
that
a
ssign
or
delega
te
com
pu
ta
ti
on
[10]
.
Hen
ce
, b
a
sed o
n
these
overall
ele
m
ents, ed
ge
co
m
pu
ti
ng
delivers
lo
w
-
la
te
nc
y, band
width
-
eff
ic
ie
ncy
,
and
r
esi
li
ent
end
-
us
er
ser
vic
es.
Using
this
serv
ic
e,
us
er
s
receive
la
te
ncy
ben
e
fits
fr
om
tho
se
w
ho
a
re
away
from
the
data
centres.
E
dge
c
om
pu
ti
ng
pro
vi
des
a
tradit
io
na
l
data
centre
with
exte
nd
e
d
capab
il
it
ie
s
to
dep
l
oy
app
li
cat
io
ns
at
the
edg
e
netw
orks[1
1].
The
pro
xim
it
y
of
us
ers
an
d
low
la
te
ncy
are
signi
ficant
adv
a
nta
ges
in
tim
es
of
netwo
r
k
co
ngest
io
n.
Th
ese
feat
ur
es
al
lo
w
th
e
m
ob
il
e
network
op
e
rato
r
s,
vend
or
s
,
as
well
a
s
app
li
cat
io
n
ser
vice
pro
vi
der
s
,
to
im
pr
ov
e
ex
ist
ing
ser
vices
us
in
g
e
dg
e
co
m
pu
ti
ng
,
e
nab
l
ing
si
gn
i
ficant
valu
e
add
it
io
n
to
the
resp
ect
ive
bus
iness
m
od
el
s.
Fi
gure
1
s
how
s
edg
e
ce
ntre
a
rch
it
ect
ure
thr
ough
LA
N
a
nd
WAN
netw
orks
c
onne
ct
ed
to t
he
cl
oud data ce
ntre
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
201
8
:
4646
-
4653
4648
Figure
1. Ed
ge
centre
arc
hitec
ture
i
n
L
AN
/
W
AN n
et
wor
k
2.1.1
.
M
od
el
of
an
Ed
ge Sys
t
em
Ed
ge
com
pu
t
ing
s
upports
pr
im
itive
virtu
al
iz
at
ion
instea
d
of
h
yper
visor
-
base
d
ha
rdwa
re
virtu
al
iz
at
ion.
In
the
pa
per
by
Saty
anar
ay
ana
n
[
12
]
,
e
dge
co
m
pu
ti
ng
is
disc
us
se
d
as
a
ne
w
par
a
dig
m
in
wh
ic
h
su
bst
antia
l
c
om
pu
ti
ng
and
st
or
a
ge
res
ource
s
are
done
at
m
ic
ro
data
cen
tres
placed
at
the
In
te
rn
et
’s
edg
e
in
cl
os
e
pro
xim
i
t
y
to
m
ob
il
e
de
vices
or
sen
s
or
s
.
The
WiCl
oud
arc
hitec
tu
re
by
H
ongxin
g
[
13
]
co
ns
ist
s
of
a
la
ye
red
a
rch
it
e
ct
ur
e
t
hat
incl
udes
a
physi
cal
la
ye
r,
vi
rtual
l
ay
er,
a
nd
a
f
unct
ion
a
nd
se
rv
i
ce
la
ye
r.
T
he
pa
per
by
Chan
g
et
al
.
[14]
giv
e
s
a
gen
eral
m
od
el
of
an
ed
ge
cl
oud
that
can
wor
k
with
al
l
ty
pe
s
of
ed
ge
ser
vices
an
d
al
so
IoT
platf
orm
s.
2.
1.2.
Appl
ic
ati
on
s
Base
d
on Edge
Com
p
ut
ing
The
Fo
ll
ow
Me
Ed
ge
(FM
E)
arc
hitec
ture
is
an
e
dg
e
ser
vice
arch
it
ect
ure
pro
posed
by
Du
tt
a
et
al
.
[15]
,
wh
e
re
the
serv
ic
e
co
ntin
uous
ly
fo
ll
ows
the
us
er
to
th
e
cl
os
est
edg
e.
Mi
gr
at
ion
is
done
t
o
ens
ur
e
t
hat
no
data
is
lo
st.
To
reali
ze
the
FME
arc
hite
ct
ur
e,
the
e
dg
e
oper
at
or
nee
ds
to
kee
p
up
dated
inf
or
m
at
ion
abo
ut
res
ources
and
us
er
locat
ion
s
.
The
SL
A
co
ns
ist
s
of
an
integrate
d
arch
it
ect
ure
of
edg
e
op
e
rato
rs,
a
sha
red
sto
ra
ge
con
ce
pt,
an
d
se
rv
ic
e
m
igrati
on
,
w
hich
are
e
nfor
ce
d
in
the
FME
arch
it
ect
ur
e
.
A
fr
am
ewo
r
k
for
m
ob
il
e
edg
e
c
om
pu
ti
ng
t
o
s
uppo
rt
div
e
rse
a
pp
li
cat
io
ns
in
a
Sm
art
ci
t
y
scenari
o,
by
re
duci
ng
cor
e
netw
ork
traf
fic
throu
gh
Sm
art
MEC
,
is
the
ov
e
rall
idea.
Anothe
r
ope
n
sens
or
platfo
rm
dev
el
ope
d
on
th
e
basis
of
e
dg
e
com
pu
ti
ng
is
pro
posed
in
Wa
gg
le
,
a
wireles
s
progr
am
that
us
es
sens
or
s
t
o
m
easur
e
ai
r
bor
ne
po
ll
uta
nts.
It
is
an
Op
e
nS
ta
c
k
-
base
d,
ed
ge
pl
at
fo
rm
that
con
sist
s
of
a
node
co
ntro
ll
er
t
ha
t
m
anag
es
a
s
ens
or
data cac
he,
rea
ds
sim
ple sen
s
or v
al
ues,
a
nd
m
anag
es t
he n
et
work stac
k
a
nd en
c
ryptio
n.
The
e
dge
c
ompu
ti
ng
f
or
the
sens
or
platf
orm
add
resses
re
sil
ie
nce,
pe
rfo
r
m
ance,
isolat
i
on,
a
nd
data
pr
i
vacy.
Ne
bu
l
a
[10]
is
a
dispe
rsed
cl
ou
d
inf
rastr
uctur
e
that
us
es
ed
ge.
It
was
de
velo
pe
d
by
the
Un
ive
r
sit
y
of
Mi
nn
es
ota
to
su
pp
or
t
dist
rib
uted
data
-
inte
nsi
ve
com
pu
ti
ng
an
d
f
or
e
ff
i
ci
ent
m
ov
em
e
nt
an
d
avail
ab
il
ity
of
la
rg
e
qua
ntit
ies
of
data
to
co
m
pu
te
reso
urc
es.
It
act
s
as
a
decen
tral
iz
ed
c
loud,
w
orki
ng
at
the
edg
e
of
the
netw
ork
a
nd
he
lping
with
the
de
vices
la
cki
ng
re
sour
ces.
F
e
m
to
Cl
oud,
di
scusse
d
in
w
or
k
by
Amm
ar
et
.
Al.
[16]
are
a
c
ollec
ti
on
of
co
-
lo
cat
ed
de
vices
us
e
d
to
pro
vide
a
cl
ou
d
se
rvi
ce
at
the
e
dge.
T
he
Fem
to
Cl
ou
d
pro
vid
es
a
dynam
ic
,
sel
f
-
conf
igurin
g,
an
d
m
ulti
-
de
vice
m
ob
il
e
cl
ou
d
ou
t
of
a
cl
us
te
r
of
m
ob
il
e
dev
ic
es.
The
arch
it
ect
ure
is
designed
to
enab
le
m
ulti
pl
e
m
ob
il
e
dev
ic
es
to
be
config
ur
e
d
into
a
coo
r
di
nated
cl
oud
com
pu
ti
ng
ser
vice.
F
r
om
the
stu
dy
on
e
dge
c
om
pu
ti
ng
and
it
s
ap
plica
ti
on
s,
it
is
se
en
that
m
igrati
on
of
resou
rces
has p
rovide
d
a
ppli
cat
ion
-
a
war
e
pr
ov
isi
on
i
ng in
e
dge
-
based ap
pl
ic
at
ion
s.
2.2.
Fo
g
Co
m
put
i
ng
Re
view
Fog
com
pu
ti
ng
was
first
int
rod
uced
by
Ci
sco
Syst
em
s,
In
c
.,
f
or
wirel
ess
data
tra
ns
f
er
bet
wee
n
distrib
uted
de
vices
in
an
I
oT
netw
ork
pa
ra
di
gm
.
Fo
g
com
pu
ti
ng,
or
f
og,
is
an
arch
it
ec
ture
that
use
s
on
e
or
m
or
e
coll
abo
ra
ti
ve
end
-
us
er
c
li
ents
or
near
-
us
er
e
dge
de
vi
ces
to
carry
out
storag
e
,
com
m
un
ic
at
ion
,
co
ntr
ol,
config
ur
at
io
n
,
m
easur
em
ent,
and
m
anag
em
ent.
Ma
ny
res
earch
pa
pe
rs
consi
der
f
og
a
nd
ed
ge
com
pu
ti
ng
com
ple
m
entary
to
on
e
a
noth
er.
Sim
il
ar
to
edg
e
c
om
pu
ti
ng,
fog
c
om
pu
ti
ng
is
ca
rr
ie
d
out
cl
os
er
t
o
th
e
end
us
er
’s
netw
ork
s.
It
is
al
so
a
vi
rtuali
zed
platf
or
m
locat
ed
be
tween
e
nd
us
e
r
s
an
d
cl
oud
da
ta
centres
ho
st
ed
on
the
I
nter
net.
It
ena
bles
c
om
pu
ti
ng
at
the
e
dge
of
the
net
w
ork
[
17]
.
E
dge
routers
are
ad
ve
rtise
d
for
pro
cesso
r
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
A Co
mparis
on
of Clo
ud Exec
ution Mec
hani
sms
F
og, E
dge
, and
…
(
T.
Fr
an
ci
s
)
4649
sp
ee
d,
nu
m
ber
of
c
or
e
s,
a
nd
buil
t
-
in n
et
work stor
a
ge,
a
nd
a
r
e u
se
d
as fo
g
s
erv
e
rs.
T
he
re is also
a cel
lula
r
base
sta
ti
on
,
a
nd
a
W
i
-
Fi
acce
s
s
po
i
nt
or
fe
m
to
cel
l
ro
ut
er
as
a
ser
ve
r.
I
n
f
og
co
m
pu
ti
ng
,
facil
it
ie
s
or
infr
a
struct
ur
es
that
ca
n
pro
vi
de
re
source
s
for
se
rv
ic
es
at
the
e
dge
of
the
netw
ork
a
r
e
cal
le
d
“
fog
nodes
.”
Figure
2
dem
on
strat
es
a
three
-
la
ye
r
use
r/
fog/cl
oud
netw
or
k.
Fog
c
om
pu
ti
ng
pro
vid
es
a
dv
a
ntage
s
in
te
rm
s
of
reducti
on in
d
e
la
y, powe
r
c
onsu
m
ption
, a
nd
reduces
data tr
aff
ic
ov
e
r
t
he n
et
work.
Figure
2. Th
re
e
-
la
ye
r
use
r/f
og/c
loud
2.2.1
.
Fo
g C
om
put
in
g Are
as an
d
Ap
pli
ca
tion
s
Fog
com
pu
ti
ng
is
us
e
d
as
op
e
n
arc
hitec
tures
for
de
vel
op
i
ng
Sm
art
li
vin
g
e
nviro
nm
ents
with
po
te
ntial
ly
thousa
nd
s
of
ve
ndors.
T
he
O
pe
n
Fog
Co
ns
ort
ium
[18]
is
a
global
syst
e
m
in
colla
borati
on
with
ARM,
Ci
sco,
Dell
,
In
te
l,
Mi
cro
s
oft
,
et
c.,
to
acce
le
rate
the
adoptio
n
of
f
og
com
pu
ti
ng
and
t
o
buil
d
a
com
m
on
ref
e
ren
ce
arc
hi
te
ct
ur
e
co
ver
i
ng
hard
war
e
a
nd
software
platfor
m
s
and
highly
so
phist
ic
at
ed
capa
bili
ti
es.
The
fog
netw
ork
connects
e
very
com
po
ne
nt
of
the
fog.
Em
erg
ing
te
chn
i
qu
e
s,
s
uc
h
as
s
oft
war
e
-
de
fin
e
d
netw
orkin
g
(SDN
)
an
d
net
w
ork
f
un
ct
i
on
vi
rtuali
zat
ion
(
NFV),
a
re
pro
po
s
ed
to
c
reate
flexible
an
d
easy
to
m
ai
ntain n
et
w
ork
e
nviro
nm
e
nts.
The
em
plo
ym
e
nt
of
S
DN
an
d
NF
V
can
ease
the
i
m
ple
m
entat
ion
an
d
m
an
agem
ent,
increase
networ
k
scal
abili
ty
,
and
red
uc
e
costs
in
m
any
aspect
s
of
f
og
c
om
pu
ti
ng,
su
c
h
as
resou
rce
al
locat
ion
,
VM
m
igrati
on,
traff
ic
m
on
it
ori
ng
,
ap
plica
ti
on
-
a
war
e
co
ntr
ol
,
an
d
program
m
able
interfac
es
[19]
.
Augm
e
nted
reali
ty
an
d
real
-
tim
e
vid
eo
ana
ly
ti
cs
are
on
e
of
the
a
reas
s
uppo
rted
by
f
og
com
pu
ti
ng
,
w
hich
ca
n
m
axi
m
iz
e
thro
ughp
ut
an
d
reduce
la
te
ncy
in
both
proces
sing
a
nd
tran
s
m
issi
on
f
or
hig
h
st
ream
vid
eo
stream
ing
pr
ocesses
.
The
I
nter
net
of
T
hings
a
nd
Connect
ed
Sm
art
de
vices
ha
ve
giv
e
n
way
to
fog
c
om
pu
ti
ng
pri
nci
ples,
w
hich
has
been
us
e
d
i
n
var
i
ou
s
dom
ain
s,
su
c
h
as
V
ehicular
net
w
orks,
Bo
dy
A
rea
Netw
orks
(BA
N)
,
an
d
t
he
Sm
art
Gr
i
d.
F
og
com
pu
ti
ng
al
lows
f
or
gr
eat
e
r
s
upport
a
nd
bette
r
respo
nse
tim
e
to
the
I
nter
net
of
t
hings
e
nviro
nm
ent,
it
is
su
it
able f
or r
e
a
l
-
t
i
m
e serv
ic
e r
equ
e
sts, a
nd it
sh
are
s r
e
source
s effi
ci
ently
[8]
.
2.3.
Co
m
pa
ri
s
on
b
etwee
n F
og an
d Edg
e
Com
p
ut
ing
Cl
oudlet
is
a
r
eso
ur
ce
-
rich
c
om
pu
te
r
li
ke
“
cl
oud
in
a
box,”
w
hich
is
av
ai
la
ble
for
us
e
by
nearby
m
ob
il
e
dev
ic
es.
Ba
hl
et
.
al
.
[20]
buil
t
Cl
oudlet,
ahea
d
of
fog
com
pu
ti
ng
,
bu
t
no
w
ag
re
e
with
the
co
nc
ept
of
fog
c
om
pu
ti
ng.
In
f
og
c
om
pu
ti
ng
,
in
fr
ast
r
uc
tures
a
re
pro
vid
e
d
as
re
sou
rces
f
or
s
er
vices
at
the
ed
ge
of
t
he
netw
ork.
T
hey
are
cal
le
d
fog
nodes
,
w
hich
a
re
si
m
il
ar
to
clo
udle
ts.
They
can
be
res
ourc
e
-
po
or
de
vices
,
su
c
h
as
set
-
top
-
bo
xe
s,
acce
ss
po
i
nt
s,
routers,
s
witc
hes,
base
sta
ti
on
s,
a
nd
e
nd
dev
ic
es
,
or
res
ource
-
rich
m
ac
hin
es
su
c
h
as
Cl
oudlet
and
IOx
.
This
stu
dy
rev
ie
w
s
po
wer
c
onsu
m
ption
by
cl
oud
and
fog
re
s
ources.
Ma
them
a
ti
call
y,
it
has
bee
n
pro
ven
t
hat
by
us
in
g
fog
res
ources
,
power
c
on
s
um
ption
is
m
ini
m
u
m
in
diff
e
ren
t
app
li
cat
io
ns
.
A
s
f
og
com
pu
ti
ng
is
a
relat
ively
new
co
nce
pt
in
cl
ou
d
c
om
pu
ti
ng,
the
pres
ence
of
secu
re
san
d
-
boxes
f
or
the
i
m
ple
m
entat
ion
of
fog
ap
plica
ti
on
s
bri
ngs
ab
ou
t
ne
w
chall
e
ng
e
s
in
te
rm
s
of
tr
us
t
and
pr
i
vacy.
Fog
ar
e
a
ty
pe
of
m
ini
-
cl
ou
ds
in
the
net
w
ork
f
or
inc
reasin
g
re
sour
ce
a
va
il
abili
ty
,
thu
s
by
do
i
ng
so,
is
ol
at
ion
and
sa
ndboxi
ng
m
echan
is
m
s
m
us
t
be
in
place
to
ensur
e
bid
irect
io
nal
tr
us
t
a
m
on
g
coope
rati
ng
pa
rtie
s.
Stand
a
r
dizat
ion
m
echan
ism
s
on
the
netw
ork
shou
l
d
ta
ke
pl
ace
at
the
te
rm
inal
an
d
no
de
end
of
f
og
networks
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
201
8
:
4646
-
4653
4650
T
her
e
is
al
so
a
la
ck
of
ce
ntra
l
entit
y
con
tr
olli
ng
the
f
og
;
he
nce,
it
is
diff
i
cult
to
asse
rt
if
a
giv
e
n
de
vi
ce
is
ind
ee
d host
ing
a com
po
ne
nt
.
At
the
sam
e
tim
e,
there
ar
e
m
any
op
en
-
end
e
d
iss
ues
in
ed
ge
c
om
pu
ti
ng
w
her
e
a
pp
li
cat
io
n
dep
l
oym
ent
strat
egies,
e
dg
e
node
secu
rity
,
an
d
fail
ur
e
re
cov
e
ry,
a
re
s
om
e
of
t
he
iss
ue
s
that
nee
d
f
ur
t
he
r
researc
h.
As
e
dg
e
is
in
cl
ose
prox
im
ity
to
end
-
us
ers
,
it
enab
le
s
reducti
on
of
la
te
ncy.
It
al
so
pro
vid
e
s
rad
i
o
netw
ork
ca
pabi
li
t
y,
includi
ng
acce
ss
to
net
work
in
form
ati
on
an
d
inte
gr
at
ion
with
op
e
ra
tor
netw
ork
se
rv
ic
e
s
.
These
are
s
ome
of
the
a
dvant
ages
of
usi
ng
edg
e
c
om
pu
ti
ng.
Table
1
pr
e
se
nts
a
com
par
at
ive
stud
y
of
f
og
a
nd
edg
e
cl
oud
c
om
pu
ti
ng
.
I
n
th
e
nex
t
sect
io
n
a
diff
e
ren
t
pe
rsp
ect
ive
of
distrib
uted
cl
oud
c
om
pu
ti
ng,
us
in
g
el
ast
i
c
execu
ti
on
m
echan
ism
s
betwee
n
de
vi
ces
and
a
cl
oud,
is
discusse
d.
F
ur
t
her
,
a
c
om
par
ison
on
cl
on
e
cl
oud,
f
og, a
nd edge c
om
pu
ti
ng is
pr
ese
nted
.
Table
1.
C
om
par
iso
n betwee
n Fo
g
a
nd E
dge
Com
pu
ti
ng
Fo
g
Co
m
p
u
tin
g
Edg
e Co
m
p
u
tin
g
1.
Dep
lo
y
ed
at
th
e lo
cal
p
re
m
ises
of
m
o
b
ile us
ers.
Dep
lo
y
ed
as a
tr
ad
itio
n
al data centre
with
exten
d
ed
capab
ilities.
2.
Virtualized
d
ev
ice
with
b
u
ild
-
in
d
ata
sto
rage
,
co
m
p
u
tin
g
,
an
d
co
m
m
u
n
icatio
n
f
acili
ty
.
Uses
an edg
e serve
r
si
m
il
ar
to
a tr
ad
itio
n
al data centre s
erver.
3.
Can
be adap
ted
f
rom
ex
istin
g
sy
ste
m
co
m
p
o
n
en
ts.
It
is co
m
p
le
tely
bu
ilt as new s
y
ste
m
o
r
a
m
in
i clou
d
data
centre.
4.
Energy
co
n
su
m
p
t
io
n
o
f
fog
is
less
th
an
clo
u
d
serv
ices, bu
t ov
erhead
is hig
h
co
m
p
a
r
ed
to clo
u
d
.
Edg
e
u
ses
less
res
o
u
rces
th
an
th
e
clo
u
d
an
d
in
itial
o
v
erhead
to
b
u
ild
is hig
h
co
m
p
ared to clo
u
d
5.
No
central e
n
tit
y
c
o
n
trollin
g
the f
o
g
clo
u
d
.
Edg
e
serv
er
u
sin
g
clo
u
d
tech
n
o
lo
g
i
es
an
d
v
irtualizatio
n
u
sed
to
co
n
trol ed
g
e co
m
p
o
n
en
ts.
6.
May
n
o
t
b
e
co
n
trolled
b
y
n
etwo
rk
o
p
erators,
u
ses
an
ad
-
h
o
c dis
tribu
tio
n
.
Allo
ws
th
e
m
o
b
ile
n
etwo
rk
o
p
erators
to
i
m
p
rov
e
ex
istin
g
serv
ices
with
edg
e.
3.
CLON
E
C
L
O
UD: ELA
STI
C
E
X
E
C
UTI
ON ME
CHA
NISM I
N
TH
E CLO
UD
The
a
dvanta
ge
s
of
cl
ou
d
c
ompu
ti
ng
a
re
t
he
serv
ic
es
offe
re
d
by
cl
oud
pro
vid
e
rs,
i.e
.
S
of
tware
-
as
-
a
-
Ser
v
ic
e
(S
aa
S),
Plat
form
-
as
-
a
-
Ser
vice
(
PaaS
),
a
nd
I
nfrastr
uc
ture
-
as
-
a
-
Ser
vi
ce
(I
aaS
).
T
o
address
t
he
in
he
ren
t
pro
blem
of
res
ource
co
ns
tr
ai
nt
in
m
ob
il
e
dev
ic
es,
the
c
oncept
of
off
-
loa
ding
data
a
nd
com
pu
ta
ti
on
to
cl
ou
d
serv
ic
e pro
vide
rs
is use
d.
Ex
a
m
ples,
suc
h
a
s
cr
owd
sourci
ng,
im
age
proc
essing, us
e o
f
G
PS
,
a
nd
I
nter
net
dat
a
happe
n
outsi
de
the
dev
ic
e
us
i
ng
m
ob
il
e
cl
oud
c
om
pu
ti
ng.
Augm
ented
Sm
artphon
e
a
pp
li
cat
ion
s,
with
cl
on
e
cl
oud,
a
ddress
chall
enges
in
ways
to
off
-
lo
ad
exec
utio
n
t
o
the
cl
ou
d
in
f
rastr
uctur
e
.
Cl
o
ne
cl
oud
te
c
hnol
og
y
was
i
ntrod
uce
d
by
Ch
un
[
21
]
for
off
-
l
oa
ding
e
xec
ution
f
ro
m
the
Sm
artph
one
to
a
com
puta
ti
on
al
infr
a
struct
ur
e
ho
sti
ng
i
n
the
cl
oud
of
Sm
artpho
ne
cl
ones.
The
idea
was
pro
po
se
d
on
th
e
si
m
ple
con
ce
pt
that
al
lows
Sm
artp
hones
t
o
host
it
s
ex
pen
sive
a
nd
e
xo
ti
c
a
pp
li
c
at
ion
s.
The
no
velty
of
t
he
a
ppr
oac
h
is
that
r
eplic
as
are lo
os
el
y sy
nc
hro
nized
a
nd
virtu
al
iz
ed
on em
ulate
d
de
vic
es.
The
f
ram
ewo
r
k
is
a
cl
oud
-
ba
sed,
fine
-
grai
ne
d,
th
rea
d
-
le
ve
l,
app
li
cat
io
n
pa
rtit
ion
er
,
w
hic
h
cl
ones
the
entire
m
ob
il
e
pla
tfo
rm
du
rin
g
runtim
e
exe
cution
int
o
th
e
cl
ou
d
Virtu
al
Ma
chine
and
r
uns
the
m
ob
il
e
app
li
cat
io
n
in
side
the
Virt
ua
l
Ma
chine,
without
pe
rform
ing
any
cha
ng
e
in
t
he
ap
plica
ti
on
co
de
.
This
appr
oach
al
so
rep
li
cat
es
the
whole
Sm
artph
one
im
age,
with
few
or
no
m
od
ifi
cat
io
ns,
int
o
powe
rful
VM
rep
li
cas,
th
us
t
ran
s
f
or
m
ing
a
sing
le
m
achine
com
pu
ta
ti
on
into
a
distrib
uted
c
om
pu
ta
ti
on
sem
i
-
autom
at
i
cal
ly
.
The
ai
m
of
off
-
loa
ding
is
al
s
o
to
m
ini
m
iz
e
the
com
m
un
ic
at
ion
a
nd
exec
ution
cost
bet
ween
the
m
ob
i
le
dev
ic
e
and
it
s
surr
oga
t
es.
Fig
ur
e
3
s
hows
a
re
plica
of
a
cl
one
cl
ou
d
-
base
d
a
rch
it
e
ct
ur
e
with
a
th
read
to
distrib
ut
e
the
process
ex
ec
ution.
Figure
3. Cl
one cloud
-
base
d archit
ect
ure
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
A Co
mparis
on
of Clo
ud Exec
ution Mec
hani
sms
F
og, E
dge
, and
…
(
T.
Fr
an
ci
s
)
4651
3.1.
Rela
ted Stu
dy on Cl
on
e
Clouds
A
stu
dy
by
Ch
ang
[
22
]
pe
rfo
r
m
ed
a
cost
ana
ly
sis
by
us
in
g
diff
e
re
nt
cl
on
e
cl
oud
res
ourc
e
al
locat
io
n
strat
egies.
The
pap
er
us
es
c
om
pu
te
r
-
inte
nsi
ve
j
ava
a
pp
li
cat
ion
s
f
or
el
ast
ic
execu
ti
on
m
echan
ism
s
in
the
cl
oud,
a
nd
t
he
y
are
com
par
ed
for
rem
ote
processi
ng
sp
ee
d,
c
os
t,
a
nd
e
ne
rg
y
sa
vings,
al
ong
-
s
ide
the
un
m
od
ifie
d
a
ppli
cat
ion
.
It
al
so
pro
ved
that
com
par
iso
ns
of
rem
ote
an
d
local
e
xecu
ti
on
s
on
net
books
a
nd
la
pto
ps
wer
e
r
edu
ce
d
f
ro
m
5%
to
50
%
,
as
com
par
ed
to
f
og
a
nd
e
dge
com
pu
ti
ng
el
ast
ic
execu
ti
ons,
wh
i
c
h
require
d
re
du
c
ed
am
ou
nts
of
cl
oud
ser
vices.
Infrast
r
ucture
dev
e
l
op
m
ent
that
is
need
e
d
in
f
og
an
d
edge
com
pu
ti
ng
is
no
t
require
d
in
cl
on
e
cl
ou
ds
.
Ela
sti
c
execu
ti
on
s
offe
r
m
or
e
flexibili
ty
to
dev
el
op
e
rs
a
nd
end
-
us
ers
to
ch
oose
cl
oud
ser
vices
at
their
prefe
r
ence.
The
a
dv
a
ntages
of u
sin
g
a
cl
one clo
ud
is
that
t
hey
la
te
-
bi
nd
the d
eci
si
on to of
f
-
l
oad exec
ut
ion
s
to
t
he
cl
ou
d
in
fr
a
struct
ur
e
, thus
giv
i
ng m
or
e
au
t
onom
y to
e
nd
-
us
ers
.
Ma
ny
fr
am
ewo
r
ks
are
de
vel
op
e
d
f
or
off
-
lo
adin
g
com
pu
ta
ti
on
s
on
cl
on
e
cl
ouds
,
s
uch
a
s
the
on
e
by
Ba
l
[2
3],
w
hic
h
pro
vid
es
a
s
yst
e
m
to
of
f
-
l
oad
m
ob
il
e
de
vice
ap
plica
ti
on
s.
T
he
ap
plica
ti
on
ge
ner
at
e
s
cod
e
from
the
or
igi
na
l
app
li
cat
ion
a
s
a
re
m
ote
servi
ce,
identic
al
t
o
the
local
one,
wh
e
re
the
rem
ote
ver
si
ons
ar
e
run
on
a
m
ulti
-
core
com
pu
te
r
in
sta
nce,
a
nd
ta
kes
f
ull
ad
vant
age
of
pa
rall
el
is
m
.
If
rem
ot
e
resou
r
ces
ar
e
not
avail
able
(s
uc
h
as
netw
ork
c
onnecti
vity
),
t
he
n
the
a
pp
li
cat
ion
ca
n
run
on
local
resou
rces
entirel
y.
The
cl
on
e
cl
oud
us
es
Vir
tual
Ma
chine
m
igrati
on
to
of
f
-
loa
d
par
t
of
t
heir
a
pp
li
cat
io
n
w
orkl
oad
t
o
a
resou
rce
-
rich
serv
e
r
thr
ough eit
he
r 3G o
r Wi
-
Fi.
The
c
os
t
m
od
el
analy
ses
the
cost
of
ex
ecuti
on
of
the
a
pp
l
ic
at
ion
on
the
dev
ic
e
a
nd
the
m
igrati
on
cost. V
irtual
M
igrati
on
re
port
s
pro
vid
e
a
n
e
ne
rg
y
sa
ving
f
r
om
60
%
to
90%
for
diff
e
ren
t
t
ypes
of
a
pp
li
ca
ti
on
s,
su
c
h
as
ches
s,
gam
es,
and
fac
e
recog
niti
on
s
yst
e
m
s.
Ther
e
are
al
so
m
any
oth
e
r
el
ast
ic
platfor
m
s
for
co
de
off
-
load,
s
uc
h
as
a
syst
e
m
le
vel
c
lon
e
an
d
deleg
at
ed
su
r
r
og
at
es
on
the
cl
ou
d.
In
al
l
these
m
e
chan
ism
s,
the
m
os
t
i
m
po
rtant r
eq
ui
rem
ent
is
the
avail
abili
ty
of
n
et
w
ork
c
onne
ct
ivit
y.
The
m
a
in
idea o
f
us
i
ng
a
cl
one
cl
oud
is
fast
execu
ti
on
an
d
m
ini
m
u
m
cost
of
se
ndin
g
data
to
t
he
cl
ou
d,
thus
si
gn
ific
a
ntly
reducin
g
the
e
xec
ution
c
os
t
on
the
dev
ic
e.
T
her
e
are
m
et
ho
ds
to
ac
hiev
e
this
thro
ug
h
analy
sis
of
con
st
raints
re
quire
d
to
par
ti
ti
on
the
app
li
cat
io
n
a
nd
cl
on
e
it
in
t
he c
loud.
Be
fore
off
-
l
oa
ding,
a
c
os
t
-
be
nef
it
a
naly
sis
s
hould
be
done
to
e
valuate
the
ben
e
fits
of
off
-
loading
an
d
the
pote
ntial
ga
in,
by
e
valuat
ing
the
predict
ed
co
st
of
exe
cution
with
use
r
-
s
pecific
re
quirem
ents.
The
cost
analy
sis
is
det
erm
ine
d
by
us
i
ng
a
prof
il
er
to
keep
trac
k
of
dev
ic
es
ene
r
gy
use
d,
netw
ork
c
har
a
ct
eris
ti
cs,
an
d
app
li
cat
io
n
cha
racteri
sti
cs. Based on
the info
rm
ation
f
r
om
t
he
pro
file
r,
the
ap
plica
ti
on
d
e
ci
des
wh
et
her
t
o
off
-
load
the
a
ppli
cat
ion
.
S
om
e
app
li
cat
ion
s
do
a
ll
ow
an
opt
i
m
i
zer
to
decide
w
hich
m
et
ho
ds
a
re
to
be
m
igrated
s
o
that
the
c
os
t
of
m
igrati
on
a
nd
e
xecu
ti
on
is
m
ini
m
iz
ed
[24].
Mo
nito
ry
cost
is
al
so
one
of
the
a
sp
ect
s
to
be
consi
der
e
d
w
hi
le
m
igrati
ng
a
pp
li
cat
io
ns
to
the
cl
oud.
Cl
one
cl
ou
d
ap
proach
e
s
ha
ve
va
rio
us
ways
o
f
us
in
g
cl
oud
res
ource
s
to
en
han
ce
t
he
com
pu
ti
ng
c
apab
il
it
ie
s
of
m
ob
il
e
dev
ic
es.
It
us
es
m
igrati
on
a
nd
re
-
i
ntegr
at
i
on
m
et
ho
ds
t
o
s
plit
m
od
ules
of
app
li
cat
io
n
bet
ween
cl
oud
a
nd
cl
one.
T
he
re
fore,
the
overa
ll
per
f
or
m
ance
an
d
cred
i
bili
ty
of
t
hese
au
gm
entat
ion
ap
proac
he
s
is
hig
hly
de
pe
nd
e
nt
on
the
cl
oud
-
base
d
re
so
urce
cha
ract
erist
ic
s.
Perfo
rm
ance,
avail
abili
ty
,
el
a
sti
ci
ty
,
vu
lnera
bili
ty
to
secur
it
y
attacks,
reli
abili
ty
,
cost,
and
distance
are
m
ajo
r
char
act
e
risti
cs
to
be
co
ns
ide
re
d
by
cl
oud
ser
vice
pr
ov
i
der
s
wh
e
n
a
ugm
ent
ing[2
5].
Ela
sti
c
exec
ution
str
at
egie
s
us
e
di
ff
e
ren
t
a
ppr
oach
es
c
ompare
d
to
fog
a
nd
e
dge
com
pu
ti
ng.
T
he
ne
xt
sect
ion
su
m
m
arizes
these
diff
e
re
nt
cl
oud
m
et
ho
do
log
ie
s.
4.
EVAL
UA
TI
O
N
O
F
DIFFE
RENT
CLO
U
D
E
X
E
C
UTI
ON MO
DELS
Fr
om
the stu
di
es p
rese
nted
, it i
s s
een th
at
all
o
f
these te
c
hnology pro
vid
e
r
s p
r
ovide d
i
ff
e
ren
t t
ypes
of
cl
oud
se
rv
ic
es
.
T
he
fi
nal
be
nef
it
s
a
re
for
the
e
nd
-
us
er
to
us
e
t
hese
ser
vices
acc
ordi
ng
to
eac
h
us
e
r’
s
requirem
ents.
Ed
ge
c
om
pu
ti
ng
an
d
fog
c
om
pu
ti
ng
are
se
tup
t
o
li
m
it
the
la
te
ncy
betw
e
en
cl
ou
d
data
centres
and
e
nd
dev
ic
es.
All
data
centres
are
co
nn
ect
ed
by
In
te
r
net
connecti
on
s
betwee
n
us
er
s
and
cl
oud
se
rv
ic
es
,
wh
ic
h
ar
e
lo
ng,
t
hin
,
an
d
s
usc
eptible
to
ne
twork
fail
ures.
In
te
rn
et
acce
s
s
is
pro
vid
e
d
by
ISPs
in
di
f
fer
e
nt
reg
i
on
s
.
Urba
n
areas
ha
ve
go
od
acce
ss
to
c
el
lular
to
wer
s
,
beca
us
e
of
th
e
de
ns
it
y
of
popula
ti
on
a
nd
hen
ce
,
In
te
r
net
c
onne
ct
ivit
y
is
al
so
avail
able
in
th
ese
areas
.
Howev
e
r,
there
ar
e
re
du
ce
d
num
ber
s
of
t
ow
e
rs
in
r
ur
a
l
areas,
t
her
e
for
e
I
ntern
et
acce
ss
an
d
acce
ss
t
o
cl
ou
d
se
rv
ic
e
s
is
m
ini
m
u
m
.
T
hu
s
,
it
can
be
a
rgue
d
that
good
acce
ss
to
cel
l
ular
towe
rs
gi
ves
good
I
nter
net
connecti
vity
,
wh
ic
h
in
tu
rn
pr
ov
i
des
good
acce
ss
to
cl
oud
pro
vid
er
s.
Fro
m
a
bu
siness
pe
rsp
ect
ive
,
the
m
or
e
data
an
d
com
pu
ta
ti
on
ob
ta
ine
d
is
of
adv
a
ntage,
as
data
i
s
charge
d
per
m
i
nu
te
t
o
c
onsum
ers.
He
nce
,
l
i
m
i
te
d
an
d
a
de
qu
at
e
us
e
of
t
he
I
nter
net
is
be
nef
ic
ia
l
to
al
l.
Cl
on
e
cl
oud
-
base
d,
distrib
uted
e
xe
cution
strat
e
gies
prov
i
de
eff
ect
ive
util
iz
at
ion
of
cl
ou
d
r
eso
urces
from
the
consum
er’
s
pe
rsp
ect
ive
.
Re
m
ote
areas
are
furthe
r
ben
e
fite
d
by
fog
a
nd
ed
ge
c
om
pu
ti
ng
due
t
o
lim
i
te
d
connecti
vity
issues.
Th
us
,
netw
ork
pro
xim
it
y
of
f
ers
a
nu
m
ber
of
a
dvanta
ges
in
te
rm
s
of
re
du
ce
d
num
ber
of
ho
ps
,
fast
respo
nd
i
ng
cl
oud
ser
vices,
et
c.,
in
th
e
cl
oud
[26]
.
I
n
urban
areas,
netw
ork
pro
xim
it
y
is
go
od
because
of
good
connecti
vity
,
wh
il
e
the
co
unte
r
is
true
in
ru
r
al
areas.
Un
li
ke
t
he
fr
e
e
su
r
rogate
re
so
urces
,
util
iz
ing
cl
ou
d
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
201
8
:
4646
-
4653
4652
infr
a
struct
ur
e
le
vies
fi
nan
ci
a
l
charges
to
t
he
e
nd
-
us
ers
.
Mob
il
e
us
ers
pay
f
or
c
on
s
um
ed
inf
rastr
uc
ture
resou
rces
acc
ordi
ng
t
o
the
S
LAs
ne
go
ti
at
ed
wit
h
cl
oud
vend
or
s
.
I
n
ce
rtai
n
sce
nar
i
os,
us
e
rs
pr
e
fer
local
execu
ti
on
or
a
pp
li
cat
io
n
te
r
m
inati
on
beca
us
e
of
t
he
m
on
et
ary
co
st
of
cl
oud
in
fr
ast
ruct
ur
es.
H
owev
er,
us
e
r
paym
ent
is
of
adv
a
ntage
t
o
cl
o
ud
ve
nd
or
s
s
o
that
they
can
m
ai
ntain
their
serv
ic
es
a
nd
de
li
ver
reli
able,
r
obus
t,
and
sec
ur
e
ser
vices
to
the
m
ob
il
e
us
e
rs
[27]
.
Th
us
,
it
ca
n
be
co
nclu
de
d
tha
t
in
ci
ti
es
an
d
i
ndus
tria
l
hubs,
cl
one
cl
oud
-
base
d
di
stribu
te
d
cl
oud
execu
ti
on
is
be
nef
ic
ia
l,
w
hil
e
s
m
al
l
town
s
and
vill
ages
shou
l
d
ta
ke
a
dv
a
ntag
e
s
of
e
dg
e
a
nd
f
og
com
pu
ti
ng.
Table
2
pro
vide
s
the
com
parat
ive
stud
y
of
al
l
of
the
di
ff
e
ren
t
cl
ou
d
exe
cution
m
echan
ism
s.
Table
2.
C
om
par
iso
n betwee
n Fo
g, E
dg
e
, a
nd Cl
one Cl
ou
d M
od
el
s
5.
CONCL
US
I
O
N
This
pap
e
r
pro
vid
es
a
n
over
vi
ew
of
the
rec
ent
de
velo
pm
e
nts
in
t
he
area
of
cl
oud
com
pu
ti
ng
a
nd
m
ob
il
e
cl
ou
d
com
pu
ti
ng.
The
pa
per
f
oc
us
es
on
th
e
rece
nt
co
nce
pts
of
f
og,
e
dg
e
,
a
nd
cl
on
e
-
base
d
com
pu
ta
ti
on
al
off
-
l
oad
i
ng
ser
vices.
Fog
c
om
pu
ti
ng
an
d
e
dg
e
com
pu
ti
n
g
are
t
he
la
te
st
in
the
de
velo
pm
ent
of
cl
oud
ser
vices,
by
br
in
gi
ng
t
he
cl
oud
se
rv
i
ces
cl
os
er
to
t
he
en
d
-
us
e
r
th
us
,
re
duci
ng
ti
m
e
of
exec
ution
on
cl
oud
ser
ver
s
.
These
te
ch
nolo
gies
are
c
om
par
ed
with
existi
ng
te
chnolo
gies
of
cl
on
e
cl
oud
-
base
d
com
pu
ta
ti
on
al
of
f
-
l
oad
i
ng
m
e
chan
ism
s.
Cl
on
e
cl
ouds
are
cat
egorized
int
o
dif
fer
e
nt
ty
pes
of
au
gm
entat
ion
strat
egies,
th
us
lim
it
ing
the
use
of
cl
oud
res
ources
.
Alto
geth
er,
these
fi
ndin
gs
in
dicat
e
ther
e
are
advanta
ge
s
in
us
in
g
cl
one
cl
oud,
fog,
a
nd
edg
e
c
om
pu
ti
ng
-
base
d
ap
pro
aches.
It
is
eviden
t
f
ro
m
the
stud
y
that
are
as
of
util
iz
at
ion
of
c
loud
res
ources
hav
e
a
n
im
pact
on
se
rv
ic
es
offer
e
d.
T
he
r
ecent
ap
proac
hes
of
f
og
an
d
edge
com
pu
ti
ng
ar
e
su
it
able
for
use
rs
with
li
m
i
t
ed
co
nnect
ivit
y
issues,
an
d
c
lon
e
cl
ou
ds
ar
e
su
it
able
f
or
op
ti
m
al
us
e
of clo
ud se
rv
ic
es.
ACKN
OWLE
DGE
MENTS
I
w
ould
li
ke
to
than
k
al
l t
he
unkn
own re
viewer
s
for t
he
c
om
m
ents an
d
s
ugge
sti
on
s
.
REFERE
NCE
S
[1]
“
A
Revi
ew
on
Eff
icient
Virtu
al
Mac
hine
Li
v
e
Migrat
ion :
Cha
l
le
nges
,
req
uir
e
m
ent
s
and
te
chnol
og
y
of
VM
.
.
.
A
Revi
ew
on
Eff
i
c
ie
nt
Virtu
al
Mac
hine
Li
v
e
Migra
ti
on :
Challeng
es
,
req
uire
m
ent
s
a
nd
te
chnol
og
y
o
f
VM
m
igra
ti
on
in
c
loud,”
Ij
-
C
lo
ser
,
no.
Ma
y
,
20
16.
[2]
N.
Ferna
ndo,
S.
W
.
Loke
,
and
W
.
Rahay
u,
“
Mobi
le
cl
oud
computi
ng:
A
survey
,
”
F
utur.
Gene
r.
Comput.
Syst.
,
vol
.
29,
no
.
1
,
pp
.
84
–
106,
2013
.
[3]
Y.
Bee
har
r
y
,
T.
P.
Fow
dur,
V.
Hurbungs
,
V.
B
assoo,
and
V.
Ramnara
in
-
See
toh
ul,
“
Anal
y
sing
t
ran
sportation
da
ta
with
open
sourc
e
big
da
ta
anal
y
t
ic
tool
s,”
Indone
s.
J
.
E
lectr.
Eng. Inf
orm
ati
cs
,
vol
.
5
,
no
.
2
,
pp
.
17
4
–
184,
2017
.
[4]
P.
Garc
i
a
Lopez
et
al.
,
“
Edg
e
-
centri
c
Com
puti
ng
,
”
ACM
SIGCO
MM
Comput.
C
omm
un.
Rev.
,
v
ol.
45
,
no
.
5
,
pp
.
37
–
42,
2015
.
[5]
Y.
Yu,
“
Mobil
e E
dge
Com
puti
ng
Towa
rds 5G :
Vision
,
Re
ce
nt
Progress
,
and
Ope
n
Challenge
s
,
”
p
p.
89
–
99
.
[6]
Y.
Zha
ng,
H
.
Liu,
L.
Ji
ao,
and
X.
Fu,
“
To
offloa
d
or
not
to
offloa
d
:
An
eff
icient
co
de
par
t
it
ion
al
go
rit
hm
for
m
obil
e
cl
oud
computing
,
”
2012
1st I
EEE
Int. Conf.
Clou
d
Net
work
ing, CLO
UD
NET 2012
-
Proc.
,
pp
.
80
–
8
6,
2012
.
[7]
X.
Masip
-
Bruin,
E.
Marín
-
Tord
e
ra,
G.
T
asha
kor,
A.
Jukan,
and
G.
J.
Ren,
“
Fogg
y
c
louds
and
cloud
y
fogs:
A
re
al
nee
d
for
coor
din
at
ed
m
ana
g
ement
of
fog
-
to
-
cl
oud
computing
s
y
stems
,
”
IEE
E
Wir
el
.
Comm
un.
,
vo
l.
23,
no.
5
,
pp
.
120
–
128,
2016
.
[8]
F.
Ai
-
doghm
ant
,
Z.
Cha
cz
ko
,
A.
R.
Aj
a
y
an
,
and
R.
Kl
empous
,
“
A
Revi
ew
on
Fog
Com
puti
ng
Te
chno
log
y
,
”
p
p.
1525
–
1530,
201
6.
Edge
Com
puti
ng
Fog Com
puti
ng
Clone
Clo
ud
1.
Close
to
end
-
use
r.
Close
to
end
-
use
r.
Us
es
distri
buted
m
ec
hani
sm
to
obta
in
Cloud
Servi
ce
s.
2.
La
t
ency
ben
efit
for
users
awa
y
from
da
ta
ce
ntr
es.
La
t
ency
ben
efit
for
users
awa
y
from
dat
a
ce
n
tres
.
La
t
ency
b
enefit
th
rough
ada
p
ti
ve
exe
cu
ti
on
sch
emes i
n
the c
loud
.
3.
Has
own
sec
urity
and
loa
d
bal
an
ci
ng.
Li
m
it
ed
se
cur
ity,
distri
but
ed
lo
a
d
bal
an
ci
ng.
Distribut
ed
loa
d
bal
an
ci
ng
and ex
ec
ut
ion.
4.
Form
s
a
thre
e
-
lay
er
serv
ice
m
odel
.
Form
s
a
thre
e
lay
er
servi
ce
m
odel
Form
s a
two
-
layer
servi
ce model
.
5.
No c
ost a
n
aly
sis
to
tr
ansfe
r
No c
ost a
n
aly
sis
to
tr
ansfe
r
Overhe
ad
cost
a
nd
ana
l
y
s
is
ta
ke
n
at
the
ti
m
e
of
o
ffloa
d
.
6.
Resourc
e
al
lo
cation
don
e
b
y
edge
nod
e.
Resourc
e
a
ll
oc
ation
done
b
y
fog
node.
Low
cost
for
surrogat
e
r
esourc
es
uti
lization.
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
A Co
mparis
on
of Clo
ud Exec
ution Mec
hani
sms
F
og, E
dge
, and
…
(
T.
Fr
an
ci
s
)
4653
[9]
B.
Varghe
se,
N.
W
ang,
S.
Barbhuiy
a,
P.
Kilpat
ric
k,
and
D.
S.
Nikolopoul
os,
“
Chal
le
ng
es
and
O
pportuni
ti
es
in
Edge
Com
puti
ng
,
”
pp
.
20
–
26
,
20
16.
[10]
M.
R
y
den
,
K.
Oh,
A.
Chandr
a, and J.
W
ei
ss
m
an,
“
Nebula
:
Distri
bute
d
Edge Cl
ou
d
for
Da
ta
Inte
ns
ive
Com
puti
ng
.
”
[11]
H.
Chang,
A.
Ha
ri,
S.
Mukherjee,
and
T.
V.
La
ksh
m
an,
“
Bringi
ng
the
cl
oud
to
th
e
e
dge,
”
in
2014
I
E
EE
Confe
ren
ce
on
Computer
Co
mm
unic
ati
ons W
orkshops
(
INFO
COM WKSHPS)
,
2014
,
pp
.
346
–
351.
[12]
M.
Sat
y
ana
ra
y
a
n
an,
“
The
Emerg
enc
e
of
Edge
Co
m
puti
ng,
”
Computer
(
Long.
Be
ach.
Cali
f)
.
,
vol
.
50,
no.
1,
pp.
30
–
39,
Jan
.
2017
.
[13]
H.
Li,
G.
Shou,
Y.
Hu,
and
Z.
Guo,
“
Mobile
edg
e
computing:
Progress
and
chall
enge
s,”
Proc
.
-
2
016
4th
IE
EE
In
t.
Conf.
Mob
.
C
lou
d
Comput.
S
erv
.
Eng.
Mob
ileClo
ud
2016
,
pp.
83
–
84,
2016
.
[14]
C.
Li
u
et
al
.
,
“
A
New
Dee
p
L
ea
rning
-
b
ase
d
Food
Rec
ognit
io
n
Sy
st
em
for
D
ie
t
ar
y
As
sess
m
ent
on
An
Edge
Com
puti
ng
Serv
ic
e
Infra
stru
ct
ur
e,
”
vol
.
1374
,
no
.
c
,
pp
.
1
–
13
,
20
17.
[15]
T.
Taleb
,
S.
Du
tt
a
,
A.
Ks
ent
ini
,
M.
Iqba
l,
and
H.
Flinc
k,
“
Mobile
Edge
Com
puti
ng
Potent
i
al
i
n
Making
Cit
ie
s
Sm
art
er,
”
IEEE C
omm
un.
Mag.
,
vol.
55
,
no
.
3
,
pp
.
38
–
43
,
Mar
.
20
17
.
[16]
K.
Haba
k,
M.
Am
m
ar,
K.
A.
Ha
rra
s,
and
E.
Z
eg
ura
,
“
Fem
toCl
ouds :
Le
ver
ag
ing
Mobile
Devices
to
Provide
Cloud
Servic
e
at t
h
e Ed
ge.
”
[17]
P.
P,
D.
K.
G.,
Y.
P,
M.
Venka
ta
Gane
sh,
and
V.
B,
“
Fog
Com
p
uti
ng:
Iss
ues,
Chal
l
enge
s
and
Future
Dire
c
ti
ons,
”
Int.
J. Ele
c
tr.
Co
mput.
Eng
.
,
vol
.
7,
no
.
6
,
p
.
3669
,
2017.
[18]
T.
Z.
Bru
ce
Mc
Mill
in,
“
Fog
Com
puti
ng
for
S
ma
rt
Li
v
ing,
”
Co
mputer
(Long.
Be
ach.
Cal
if
)
.
,
no.
Februa
r
y
,
p.
20
17,
2017.
[19]
S.
Yi,
C.
L
i,
and
Q.
Li
,
“
A
Survey
of
Fog
Com
puti
ng :
Conc
ept
s
,
Applic
a
ti
ons
and
Iss
ues,
”
in
Mobidat
a
,
2015
,
pp
.
37
–
42.
[20]
M.
Sat
y
an
aray
a
nan,
P.
Bahl,
R.
Các
er
es,
and
N.
Davie
s,
“
The
c
a
se
for
VM
-
base
d
cl
oudl
et
s
in
m
obil
e
computing,”
IEE
E
Pe
r
vasiv
e
Compu
t.
,
vo
l. 8,
no.
4
,
pp
.
14
–
23
,
2009.
[21]
B.
-
G.
Chun,
S.
I
hm
,
P.
Mania
t
is,
M.
Naik,
and
A
.
Patti,
“
Clone
C
l
oud,
”
in
Procee
dings
of
the
sixth
conf
ere
n
ce
on
Computer
syste
ms
-
EuroSys
’1
1
,
2011
,
p
.
301
.
[22]
Y.
Chang,
S.
Hu
ng,
N.
J.
C.
W
a
ng,
and
B.
Li
n,
“
CS
R :
a
Cloud
-
assisted
Speec
h
Rec
ognition
Ser
vic
e
for
Persona
l
Mobile
Dev
ice,
”
in
2011
Int
ernat
ional
Conf
ere
nc
e
on
Parallel Proce
ss
ing
,
2011
,
pp.
305
–
314
.
[23]
R.
Kem
p,
N.
Pal
m
er,
T.
Ki
el
m
an
n,
and
H.
Ba
l,
“
Cuckoo :
a
Com
puta
ti
on
Offlo
ad
ing
Fram
ework
for
Sm
art
ph
ones,
”
in
Se
cond
In
te
rn
ati
onal ICST
Co
nfe
renc
e,
Mob
iC
ASE
2010
.
[24]
M.
Shiraz,
S.
Abolfa
zli,
Z.
S
ana
e
i,
and
A.
Gani,
“
A
stud
y
on
virt
ua
l
m
ac
hine
d
epl
o
y
m
e
nt
for
applic
at
i
on
outsourci
ng
in
m
obil
e
cl
oud
com
puti
ng,
”
J. Supe
r
comput.
,
vol
.
63
,
no.
3,
pp.
946
–
9
64,
2013
.
[25]
J.
Li
u
,
E
.
Ahm
ed,
M.
Shir
az
,
A.
Gani,
R
.
Bu
yy
a
,
and
A.
Qur
eshi,
“
Journal
of
Net
work
and
Com
pute
r
Appli
cation
s
Applic
a
ti
on
par
t
it
ioni
ng
al
gor
it
h
m
s
in
m
obil
e
cloud
computing :
Ta
xonom
y
,
r
e
vie
w
and
fu
ture
dire
c
ti
ons,
”
J
.
Net
w.
Comput.
Appl
.
,
vo
l.
48,
p
p.
99
–
117
,
2015
.
[26]
H.
J.
La
and
S.
D.
Kim
,
“A
ta
xonom
y
of
offloa
d
ing
in
m
obil
e
cloud
computing,
”
Proc.
-
IEE
E
7t
h
Int.
Conf.
Serv
.
Comput.
App
l. S
OCA
2014
,
pp.
1
47
–
153,
2014
.
[27]
S.
Abolfa
z
li,
Z
.
Sanaei,
E.
Ahm
ed,
A.
Gan
i,
and
R.
Bu
yy
a,
“
Cloud
-
Ba
sed
Augm
ent
at
ion
for
Mobile
D
evi
c
e
s:
Motiva
ti
on
,
Ta
x
onom
ie
s,
and
Open
Cha
ll
eng
es,
”
pp.
1
–
32
,
2013
.
Evaluation Warning : The document was created with Spire.PDF for Python.