Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
9
, No
.
5
,
Octo
ber
201
9
, pp.
4433
~
44
40
IS
S
N:
20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v
9
i
5
.
pp
4433
-
44
40
4433
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
An
over
view
of
virtual m
ac
hin
e live m
igrati
on tech
niq
ues
Artan M
az
rekaj
1
,
Shkel
z
en N
uz
a
2
, Mi
moz
a
Z
at
ri
qi
3
, Vle
ra Ali
meha
j
3
1
Facul
t
y
of
Cont
emporar
y
Sci
ence
and
T
ec
hnolog
ie
s,
South
E
ast
E
urope
an
Univ
ers
ity
,
Republic
of N
orth
Mac
edon
i
a
2
Instit
ute of
N
atural
and
Appl
ie
d
Scie
n
ce
s,
Dokuz
E
y
l
ul
Univ
ersi
t
y
,
Turkey
3
Facul
t
y
of Elect
ric
a
l
and
Com
pute
r Engineering
,
Univer
sit
y
of
Pri
shtina
,
Kos
ovo
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
ul
18
, 2
01
8
Re
vised
A
pr
2
0
, 2
01
9
Accepte
d
Apr 30
, 201
9
In
a
cl
oud
computing
the
l
ive
m
igra
ti
on
of
virt
u
a
l
m
ac
hine
s
show
s
a
proc
ess
of
m
oving
a
ru
nning
virt
u
al
m
ac
hin
e
from
source
ph
y
si
ca
l
m
ac
hine
to
th
e
desti
nation,
con
sideri
ng
the
CP
U,
m
emor
y
,
ne
twork,
and
stor
age
st
at
es
.
Vari
ous
per
form
anc
e
m
e
tri
cs
are
ta
ck
le
d
such
as,
downtime,
to
ta
l
m
igra
ti
o
n
ti
m
e,
per
form
an
ce
degr
ada
t
ion,
and
amount
of
m
igra
te
d
data,
which
ar
e
aff
ecte
d
when
a
virt
ual
m
ac
hin
e
i
s
m
igra
te
d.
Thi
s pa
per
pr
ese
nts
a
n
over
vie
w
and
under
standing
of
virt
ual
m
ac
hine
li
v
e
m
igra
ti
on
t
ec
hn
iq
ues,
of
the
diffe
ren
t
works
i
n
li
t
era
tur
e
th
at
conside
r
th
is
issue,
which
m
ight
impact
th
e
work
of
profe
ss
iona
ls
and
rese
ar
che
rs
to
furth
er
expl
ore
the
ch
allenge
s
and
provide
op
ti
m
al
soluti
ons.
Ke
yw
or
d
s
:
Cl
oud
c
om
pu
ti
ng
Live m
igrati
on
Po
st
-
C
opy ap
proac
h
Pr
e
-
C
opy ap
proach
Virtuali
zat
ion
Copyright
©
201
9
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
:
Shkelz
en N
uza
,
In
sti
tute
of N
at
ur
al
a
nd
Applie
d
Scie
nces
,
Dok
uz
Ey
lul
U
niv
e
rsity
,
Bou
le
vard: D
E
Ü
Fe
n
Bi
li
m
le
ri En
sti
tüs
ü
Co
ğr
a
fi Bi
lgi Sist
e
m
le
ri A
na
bili
m
D
al
ı Tı
nazte
pe Kam
pü
sü
,
35160 B
uca/İ
Z
MİR, T
urkey
.
Em
a
il
:
sh
kelzen.
nu
za
@
gm
ail.co
m
1.
INTROD
U
CTION
Cl
oud
c
om
pu
ti
ng
is
known
a
n
em
erg
in
g
pa
rad
i
gm
wh
ere
so
ft
war
e
,
platfor
m
and
inf
ras
tructu
re
ca
n
be
acce
sse
d
as
a
serv
ic
e
.
I
n
the
cl
oud
com
pu
ti
ng
t
he
key
con
ce
pt
is
Vi
rtuali
zat
ion
that
al
lows
s
har
i
ng
of
a
sing
le
i
ns
ta
nce
of
an
ap
plica
ti
on
or
a
re
sour
ce
betwee
n
m
ulti
ple
custom
ers
an
d
orga
nizat
ion
s.
Virt
ualiz
at
ion
create
s
a
virtu
a
l
env
ir
on
m
ent
on
a
sin
gle
ph
y
sic
al
m
achine
by
abstracti
ng
the
hard
war
e
de
ta
il
s.
It
allow
s
us
to
us
e
m
ulti
ple
i
ns
ta
nces
of
operati
ng
syst
e
m
s
(k
no
wn
as
gu
est
OS)
to
handle
the
num
ber
of
pro
cesses
si
m
ultaneou
sly
and
sepa
ratel
y
by
each
gues
t
op
e
rati
ng
syst
e
m
[
1
]
.
It
assi
gn
s
a
lo
gical
nam
e
cor
res
pond
to
a
ph
ysi
cal
sto
ra
ge
an
d
prov
i
des
a
pointer
to
t
ha
t
ph
ysi
cal
re
s
ource
wh
e
n
it
i
s
dem
and
e
d.
T
her
e
are
fou
r
know
n
ty
pes
of
virtu
al
iz
at
ion
s:
1.
Hardwa
re
virt
ualiz
at
ion
:
w
he
n
the
virt
ual
m
achine
(VM
)
softwa
re
or
the
VM
m
a
nag
e
r
is
i
ns
ta
ll
ed
directl
y o
n
the
hard
war
e
syst
em
.
2.
Op
e
rati
ng
syst
e
m
virtu
al
iz
at
ion
:
wh
e
n
t
he
VM
softwa
re
or
the
VM
m
anag
er
is
instal
le
d
on
t
he
host
OS
and not
directl
y on the
ha
rdw
are syst
em
.
3.
Ser
ver
virt
ualiz
at
ion
:
wh
e
n
t
he
VM
s
of
t
wa
r
e
or
t
he
VM
m
anag
er
is
i
nst
al
le
d
directl
y
on
the
se
rv
e
r
syst
e
m
.
4.
Stor
a
ge
Virtua
li
zat
ion
:
gro
uping
process
of
the
physi
cal
stora
ge
from
different
m
ulti
ple
netw
ork
st
or
a
ge
dev
ic
es
to
l
ook l
ike a si
ng
le
stora
ge de
vice.
As
virt
ualiz
at
i
on
s
plit
s
a
ph
y
sic
al
m
achi
ne
(P
M)
into
se
ve
ral
VMs
,
t
his
br
i
ng
s
us
t
o
th
e
te
rm
of
t
he
VM
w
hich
is
a
softwa
re
im
plem
entat
ion
of
an
en
vir
onm
ent
of
c
om
pu
ti
ng
w
here
a
program
or
OS
can
be
instal
le
d
an
d
r
un
[
2
]
.
VM
wa
re
ES
X
/
E
SX
i
[
3
]
,
Virt
ual
P
C
[
4
]
,
Xe
n
[
5
]
,
an
d
Mi
croso
ft
Hype
r
-
V
[
6
]
,
KV
M
[
7],
VirtualBo
x
[
8
]
know
n
as
hype
rv
is
or
s
w
hich
a
re
so
m
e
popula
r
virtu
al
iz
at
ion
softwa
r
e.
Xe
n
an
d
V
Mware
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.
9
, N
o.
5
,
Oct
ober
201
9
:
4
4
3
3
-
4
4
4
0
4434
hype
rv
is
or
s
ha
ve
s
pecial
te
chnolo
gy
f
or
li
ve
m
igrati
on
,
and
t
hey
are
know
n
as
XenM
otion
a
nd
V
Moti
on
.
Feng
et
al
.
[
9
]
com
par
e
the
pe
rfor
m
ances
of
both
an
d
t
he
giv
e
n
res
ults
wh
e
n
m
igrati
ng
id
entic
al
V
Ms
sho
w
that
VMotio
n
gen
e
rates
le
ss
data
trans
ferre
d
than
Xe
nMot
ion
.
Live
m
igrati
on
of
VMs
of
fe
rs
the
po
s
sibil
it
y
for
al
locat
io
n
of
resou
rces
to
runn
i
ng
ser
vic
es
with
ou
t
int
e
rrup
ti
on
durin
g
m
igrati
on
pro
cess
that
is
i
m
portant
for
se
rv
ic
es
w
i
th p
a
rtic
ular
Quali
ty
o
f Ser
vi
ce (
QoS) re
qu
i
rem
ents
[10
].
Back
gr
ou
nd
Live
m
igrati
on
of
VMs
is
a
process
of
m
igrati
ng
the
sta
te
s
(
CPU,
m
e
m
or
y,
storag
e
,
net
work,
et
c.)
of
VM
f
ro
m
on
e
ph
ysi
cal
m
achine
to
a
no
t
her
on
e
.
On
t
he
pe
rfor
m
ance
of
li
ve
m
igrati
on,
var
i
ous
te
ch
ni
qu
e
s
m
ake v
a
rio
us
i
m
pacts. I
n gene
ral, li
ve
m
igra
ti
on
of V
Ms
h
a
s sev
e
ral
ben
e
f
it
s includ
i
ng
:
1.
On
li
ne
ma
i
nte
nance,
s
om
et
i
m
es
to
enh
anc
e
reli
abili
ty
an
d
avail
abili
ty
of
a
syst
e
m
,
i
t
m
us
t
be
con
ne
ct
ed
to cli
ents s
o
al
l
V
Ms a
re m
igrati
ng
a
way wit
hout
bein
g disc
onnected
.
2.
Load
Bal
an
ci
ng
,
w
hen
t
he
lo
ad
is
c
onsidera
bly
unbala
nce
d,
t
he
VMs
s
houl
d
be
m
igrated
f
r
om
PMs
t
hat
are
ov
e
rloa
de
d t
o othe
r
PMs
that are
not in
the
ov
e
rloa
de
d
sta
te
.
3.
Ma
nage
ab
il
it
y a
nd
m
ain
te
na
nc
e
, m
ov
em
ents
of V
Ms
and
s
hu
t
dow
n of P
Ms for m
ai
ntenan
ce
.
4.
Ener
gy
Ma
nage
men
t
,
c
onso
li
dation
of
VMs,
switc
h
off
unde
r
util
iz
ed
PMs
to
red
uc
e
data
center
’s
he
at
loss a
nd po
wer co
nsum
ption
.
5.
Impr
oved pe
rforma
nce
and r
el
iab
il
it
y
,
the a
pp
li
cat
io
n per
f
or
m
ance w
il
l n
ot b
e
d
e
grade
d.
6.
Mi
nimum
vi
ol
ation
of
Service
Level
Agreem
ent
(
SL
A)
,
meeti
ng
the
SLA
req
uireme
nts
betwe
en
cl
ou
d
pr
ov
iders
an
d cl
oud users
.
Ther
e
a
re
m
an
y
pap
ers
that h
and
le
s
the V
M
m
igrati
on
te
ch
niques, b
ut
this
pap
e
r
is
i
nten
ded
to
m
ake
a b
et
te
r u
nd
e
rs
ta
nd
in
g o
f VM m
igrati
on tech
niques.
In
[
11
]
,
on
di
ff
e
ren
t
hype
r
vi
so
rs
are
com
par
e
d
li
ve
m
i
gr
at
io
n
e
ff
ic
ie
ncies.
Xe
n
s
pe
nt
the
m
os
t
dow
nti
m
e w
hile K
VM s
pen
t t
he
le
ast
s
um
o
f
downti
m
e fo
r st
or
a
ge
a
nd m
e
m
or
y l
ive
m
igrati
on
s.
In
[
12
]
,
is
re
presented
pr
e
-
c
opy
a
pproach
t
ha
t
sh
ows
bette
r
pe
rfo
rm
ance
com
par
ed
with
pure
st
op
-
and
-
co
py
.
T
he
pr
e
-
c
op
y
com
bin
es
m
any
ro
unds
of
pus
h
and
s
hort
stop
-
and
-
co
py
at
the
end
.
T
he
sto
p
-
and
-
cop
y
has
high
dow
nti
m
e.
The
pr
e
-
c
opy
te
ch
nique
is
bette
r
than
on
-
dema
nd
,
because
sto
p
-
and
-
co
py
ha
s
high
total
m
igrati
on
tim
e,
and
c
om
pu
te
r
res
our
ces
in
strai
gh
t
pro
portio
n
w
it
h
increasi
ng
the
tim
e.
The
m
os
t
do
m
inant para
m
et
er o
n pe
rfo
rm
ance is the
m
igrati
on
li
nk
sp
ee
d.
Also
,
i
n
[
13
]
are
su
m
m
arize
d
the
a
dv
a
ntag
es
of
dif
fer
e
nt
appro
a
ches
,
s
uch
as
pr
e
-
co
py
an
d
sto
p
-
and
-
co
py
.
T
he
per
f
or
m
ance
of
li
ve
m
igrati
on
can
be
af
fected
dif
fer
e
nt
ly
us
ing
trad
e
-
off
te
ch
niqu
es
on
diff
e
re
nt
hype
rv
is
or
s
.
Using
diff
ere
nt
virt
ualiz
at
ion
te
chn
i
qu
e
s
on
di
ff
ere
nt
hype
r
visors,
aut
hor
s
has
co
m
par
ed,
CP
U
us
a
ges,
m
e
m
or
y
utilizat
io
ns
an
d
trans
fer
tim
e.
In
[14
]
,
the
auth
or
s
s
how
that
ap
plyi
ng
pr
e
-
paging
te
c
hn
i
que
im
pr
oves
th
e
pa
ge
fa
ult
pr
ob
le
m
.
W
or
king
set
s
t
hat
will
be
us
ed
in
t
he
fu
t
ur
e
a
re
pr
e
di
ct
ed,
and
pa
ges
a
re
loade
d
be
f
or
e
bein
g
acce
sse
d,
then
norm
al
po
st
-
co
py
li
ve
m
igrati
on
is
processe
d.
In
or
der
t
o
el
i
m
inate
the
m
igrati
on
of
free
m
e
m
or
y
pag
es,
a
m
echan
ism
dyn
amic
sel
f
-
ba
ll
oo
ning
will
be
execu
t
ed.
T
he
resu
lt
s
from
pre
-
co
py
on
X
en
are
com
par
ed
with
the
im
pr
ov
e
d
post
-
c
op
y
ap
proac
h
a
nd
it
s
hows
that
post
-
cop
y
is
bette
r
in
connecti
on
with
num
ber
of
trans
ferred
pa
ges
an
d
total
m
igrati
on
tim
e
,
wh
il
e
pr
e
-
co
py
is
bette
r
in
term
s o
f
downti
m
e.
In
orde
r
t
o
m
ake
t
he
op
ti
m
al
sel
ect
ion
for
destinat
io
n
VMs,
i
n
[15
]
is
use
d
a
c
ost
m
od
el
on
hype
rv
is
or
X
e
n.
F
or
t
he
cal
culat
ion
,
t
he
dif
fer
e
nt
param
eter
s
are u
sed
,
suc
h
is
the
su
m
of
netw
ork
tra
ff
i
c,
siz
e
of
VM
’s
m
e
mo
ry
an
d
de
gr
ee
of
dirty
pa
ges.
Du
ri
ng
the
c
re
at
ion
of
a
cost
-
awar
e
m
igrati
on
al
gorithm
[1
6
]
us
e
a
Sh
a
non’s
no
t
ion
of
e
ntr
opy
[16
]
.
By
cal
cul
at
ing
the
t
rad
e
off
of
perform
ance
in
flue
nce
and
m
igrati
on
tim
e
the alg
or
it
hm
w
ants
to
m
inim
iz
e
m
igrati
on
co
st.
In
[
17
]
,
Live
Gang
Mi
gr
at
ion
ex
plains
,
wh
e
re
t
he
gro
up
of
VMs
m
igrates
sim
u
lt
aneo
usl
y
by
QEMU/
KV
M
hype
rv
is
or
s
w
i
th a m
ini
m
u
m
t
otal netw
ork
tr
aff
ic
a
nd m
igrati
on
ti
m
e. I
n order
t
o do co
nc
urren
t
li
ve
m
igrati
on
of
co
-
locat
e
d
VMs
is
us
ed
D
e
-
du
plica
ti
on
te
chn
i
qu
e
.
De
-
duplica
ti
on
de
al
s
with
el
i
m
inati
on
of
redu
nd
a
nt
in
form
ation
.
In
order
t
o
detect
s
i
m
i
la
rity
of
co
ntents
pag
e
i
de
ntica
l
detect
or
us
e
s
has
h
f
unct
ion.
The
res
ult
of
t
his
pap
e
r
s
how
s
that
Liv
e
Ga
ng
Mi
gr
at
io
n
c
an
reduce
t
otal
netw
ork
traf
fic
an
d
m
igrati
on
ti
m
e.
In
orde
r
t
o
re
du
ce
the
m
igr
at
ion
loa
d
by
reducin
g
set
of
the
data
t
o
m
igrate,
sug
ge
sts
ne
w
m
od
el
nam
ed
In
c
rem
ental
. I
t uses
blo
ck
-
bitmap
i
n order
to
sync
hron
iz
e al
l wr
it
e accesse
s to
t
he
local
di
sk
[18
]
.
Proble
m Descr
ipti
on
Ther
e
ha
ve
be
en
c
on
si
der
a
ble
co
ntributi
on
s
in
the
li
ve
m
igrati
on
of
VM
te
c
hn
i
ques.
For
the
te
chn
iq
ues
of
VM
li
ve
m
igrati
on
that
ha
ve
been
pr
ese
nt
ed
so
far,
in
orde
r
to
fi
nd
a
n
op
ti
m
al
so
lutio
n
of
dynam
ic
con
so
li
dation
i
n
the
cl
oud
e
nv
iro
nm
ents,
m
a
ny
ap
prox
im
a
ti
on
te
ch
nique
s
hav
e
bee
n
us
e
d.
By
co
m
par
ing
the
m
ajo
r
te
chn
i
qu
e
s
of
V
M
li
ve
m
igrati
on
in
the
cl
oud
en
vir
on
m
ents,
it
has
enab
l
ed
us
t
o
unde
rstan
d
the
i
m
pact
factor
s
and
li
m
it
a
ti
on
s
that
em
erg
e
from
these
te
c
hn
i
qu
e
s,
th
us
m
aking
th
e
na
rrowe
r
def
i
niti
on
of
t
he
iss
ues
relat
ed
to
dyn
am
ic
co
nsoli
dation
an
d
res
ource
util
iz
at
ion
,
s
uc
h
as
a
ver
y
c
om
plex
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
An overvi
ew
of virt
ua
l
m
ach
i
ne
li
ve migr
atio
n
te
ch
niques
(
Artan M
az
rek
aj
)
4435
issue
in
cl
oud
infr
a
struct
ur
e
.
The
m
ai
n
fo
c
us
of
this
pap
e
r
is
to
pro
vid
e
to
the
researc
he
rs
com
pr
e
hensi
ve
unde
rstan
ding an
d
cl
assifi
cat
ion
of k
ey
as
pe
ct
s
of
li
ve
m
igrati
on
te
ch
niqu
es,
on
res
ource
s
li
ke
CP
U,
m
em
or
y,
netw
ork, an
d
st
or
a
ge.
Con
tri
buti
ons
In
t
his
pap
e
r,
t
he
prob
le
m
of
dynam
ic
con
s
ol
idati
on
of
VM
s
thr
ough
t
he
li
ve
m
igrati
on
m
echan
ism
is
co
ns
ide
red,
wh
ic
h
e
nab
le
s
eff
ic
ie
nt
res
ource
al
locat
ion
in
the
cl
oud
e
nv
i
ronm
ents.
The
m
ai
n
para
m
et
ers
that
hav
e
a
dir
ect
i
m
pact
on
the
eff
ic
ie
ncy
of
the
res
ourc
es
that
are
ta
ckled,
s
uc
h
as
dow
nti
m
e,
disruptio
n
tim
e,
total
m
igrati
on
ti
m
e,
a
m
ou
nt
of
m
igrated
data,
a
nd
perform
ance
de
gr
a
datio
n.
F
urt
her
m
or
e,
i
n
te
rm
s
of
li
ve
m
igrati
on
,
the
m
ajo
r
te
chn
i
qu
e
s
that
ha
ve
an
im
pact
on
res
ource
uti
li
zat
ion
are
de
al
t
with
rig
oro
us
ly
as
m
e
m
or
y, file
, net
w
ork, an
d d
evice m
igrati
on
.
This
pa
per
is
orga
nized
as
f
ol
lows
.
Sect
i
on
2
ex
plains
the
perform
ance
m
et
rics.
In
the
s
ect
ion
3
are
ta
ckled
the
li
ve
VM
m
igra
tio
n
te
c
hniq
ues,
wh
e
reas
i
n
t
he
sect
io
n
4
are
gi
ven
ty
pe
s
of
ot
her
V
M
li
ve
m
igrati
on
tec
hniq
ues. The
p
a
per co
nclu
des wit
h
a
su
m
m
ar
y and a c
om
par
ison o
f rel
at
ed work
.
2.
PERFO
R
MANC
E
METR
I
CS
The
f
ollow
i
ng
m
et
rics
are
m
os
tl
y
us
ed
to
m
easur
e
the
ef
fici
ency
and
pe
rfor
m
ance
of
a
VM
li
ve
m
igrati
on
proc
ess.
2.1.
D
owntim
e
This
m
et
ric
rep
rese
nts
th
e
in
te
rv
al
of
ti
m
e
wh
il
e
se
rv
ic
es
are
no
t
run
ni
ng
an
d
a
vaila
bl
e.
Actuall
y,
is t
he
ti
m
e d
urat
ion
from
w
he
n VM pa
us
es
on th
e
s
ource P
M t
il
l i
t resu
m
es on t
he desti
nation PM
.
2.2.
Disr
upt
i
on time
It
is
the
tim
e
wh
e
n
cl
ie
nts
that
are
co
nnec
te
d
to
the
ser
vi
ces
that
are
runn
i
ng
on
t
he
m
igrated
VM
no
ti
ce
de
gr
a
da
ti
on
of
ser
vice
respon
sive
nes
s.
The
refor
e
,
wh
e
n
cl
ie
nts
r
equ
e
st
any
ser
vice,
res
pons
e
tim
e
ta
kes
lo
nger
.
This
m
eans,
di
sruptio
n
ti
m
e
i
s
the
tim
e
per
i
od
durin
g
w
hich
ser
vices
on
the
Virtual
Ma
chine
sh
ow
lo
we
r
pe
rfor
m
ance
to
the
cl
ie
nt
beca
use
of
t
he
m
igrati
on
pr
ocess.
Al
so
the
m
et
ho
ds
for
sync
hro
ni
zat
ion
and the t
ran
s
fe
r
rates
h
a
ve
a
n i
nf
lue
nce
on th
is perf
or
m
ance m
e
tric
.
2.3.
To
ta
l
mi
gr
at
i
on
time
This
m
et
ric
rep
rese
nts
the
ti
m
e
du
rati
on
since
the
m
igrati
on
sta
rts
ti
ll
t
he
sta
te
s/servi
ces
on
the
so
urce
an
d
de
sti
nation
PM
are
to
ta
ll
y
sy
nchr
on
iz
e
d.
T
o
reduce
total
m
igrati
on
tim
e
i
t
is
pr
efe
rr
e
d
for
decr
easi
ng the
siz
e o
f
tra
ns
fe
r
red data,
for ex
a
m
ple to c
om
pr
ess t
he
tra
ns
fe
rr
e
d data be
f
ore we se
nd it
.
2.4.
Am
ount
of
mi
gr
at
e
d dat
a
It
is
the
a
m
ou
nt
of
data
that
ar
e
transm
itted
dur
i
ng
the
w
ho
l
e
tim
e
of
m
igrati
on
.
S
om
et
i
mes
it
can
be
ref
e
rr
e
d
to
as pa
ges
tra
nsfer
re
d.
T
he
m
ini
m
a
l
a
m
ou
nt
is
c
onside
red
the
si
ze
of
the r
un
ti
m
e
sta
te
s,
li
ke
stora
ge
siz
e,
m
e
m
or
y
siz
e,
CPU
sta
t
e
siz
e,
et
c.
In
m
os
t
cases
it
will
be
great
er
than
the
act
ua
l
r
un
tim
e
sta
t
e
siz
e.
This
will
no
t
happe
n
in
f
re
eze
-
an
d
-
c
opy
m
et
ho
d
because
in
this
m
eth
od
m
us
t
be
so
m
e
red
unda
ncy
for
protoc
ols a
nd s
ynch
ronizat
ion.
2.5.
Perf
orm
ance
de
gradati
on
This
m
et
ric
rep
rese
nts
t
he
de
crease
of
pe
r
form
ance
of
the
se
rv
ic
e
w
hi
ch
is
ca
us
e
d
durin
g
t
he
m
igrati
on
proc
ess.
Per
form
ance
deg
ra
datio
n
is
evaluated
by
co
m
par
in
g
the
thr
oughput
of
ser
vice
duri
ng
the
m
igrati
on
a
nd
without i
t.
3.
LIVE V
M MI
GRATIO
N
T
ECHNIQ
UES
3.1.
Me
mor
y mi
gration
Mem
or
y
m
igrati
on
is
a
process
of
m
igr
at
ing
VM
m
e
m
or
y
instance
fr
om
the
so
ur
ce
to
the
destinat
io
n.
Th
e proces
s
of
m
e
m
or
y m
igratio
n ca
n be
d
i
vide
d
int
o
ph
a
ses
[19
]:
1.
Pu
s
h
ph
ase:
th
e
hype
rv
is
or
t
r
ansf
e
rs
m
e
m
or
y
pag
es
t
o
the
destinat
io
n
P
M
wh
e
n
VM
on
source
is
sti
ll
run
ning.
Dirtie
d
pa
ges
in
tra
ns
m
issi
on
proc
ess
are
sent
ag
ai
n
ti
ll
the
rate
of
those
rec
opie
d
pa
ges
are
m
or
e than
dirt
yi
ng
rate f
or co
ns
ist
ency.
2.
Stop
-
a
nd
-
co
py
phase:
a
fter
be
ing
co
pied
th
e
m
e
m
or
y
pages
by
source
V
M
w
ho
ha
s
be
en
st
oppe
d
th
a
n
sta
rt
a
n
ew
V
M
to
the
desti
nation.
T
his
m
eans
the
s
ourc
e
VM
has
st
opped
unti
l
pag
e
s
has
been
c
op
ie
d
to the dest
inati
on V
M,
and
th
en
the
n
e
w V
M has sta
rted.
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.
9
, N
o.
5
,
Oct
ober
201
9
:
4
4
3
3
-
4
4
4
0
4436
3.
Pu
ll
phase:
Af
t
er
exec
ution
of
the
VM
on
de
sti
nation
PM
s
ta
rts
if
a
pag
e
require
d
is
no
t
found,
the
n
a
pag
e
f
a
ult occ
urs, s
o
t
he pa
ge i
s pull
ed
f
r
om
t
he
s
ource
VM
thr
ough the
n
et
work.
All
the
m
igratio
n
te
ch
niques
try
to
red
uce
t
otal
m
igrati
on
tim
e
and
dow
n
tim
e.
Ther
e
ar
e
two
m
a
in
appr
oach
es
in m
e
m
or
y
m
igrati
on
s: a)
P
re
-
C
op
y a
nd
b) P
ost
-
Cop
y
.
3.1.1.
Pre
-
c
opy
ap
pr
oach
In
the
pre
-
c
opy
app
r
oac
h
al
l
m
e
m
or
y
pag
e
s
are
sent
befo
re
the
VM
is
r
esum
ed
to
exe
cute
on
the
new
node
.
Thi
s
te
chn
iq
ue
ha
s
been
im
ple
m
ented
on
m
any
kinds
of
hype
r
visors
li
ke
Xe
n,
VMwa
re
an
d
KV
M
and
us
es
it
erati
ve
pu
s
h
an
d
s
top
-
and
-
c
op
y
phas
e
[20
]
.
D
ue
to
the
it
erati
ve
proce
dure
of
pus
h
phase
so
m
e
m
e
m
or
y
pag
es
,
w
hich
a
re
known
as
dirty
pa
ges
will
be
m
od
ifie
d.
T
hen
th
ese
pa
ges
a
re
r
egen
e
rated
on
so
urc
e
PM
durin
g
it
erati
on
s
of
the
m
igrati
on
pr
oc
ess.
I
n
the
sec
ond
phase
,
te
r
m
inati
on
de
pe
nd
s
on
the
de
fine
d
thres
ho
l
d.
The
te
rm
inati
on
phase is exec
ute
d i
f
a
ny one
of c
onditi
ons m
eet
[
21
]:
a.
Nu
m
ber
of it
er
at
ion
s e
xcee
ds
the num
ber
of
i
te
rati
on
s
pre
-
de
fine
d,
b.
The
t
otal am
ou
nt of m
e
m
or
y wh
ic
h has
bee
n
se
nt, or
c.
The n
um
ber
of
total
d
irty
pages in
pr
e
vious
roun
d fall
s b
el
ow the
d
e
fine
d
t
hr
es
hold.
In
the
pu
ll
pha
se,
at
so
urce
P
M
the
VM
m
i
gr
at
in
g
proce
s
s
is
su
sp
e
nd
e
d,
and
a
fter
that
rem
a
ining
dirty
pa
ges
a
nd
proces
sor’s
sta
te
are
su
s
pende
d.
Wh
e
n
the
m
igrati
on
pr
ocess
is
com
plete
d
correct
ly
,
hype
rv
is
or r
es
um
es V
M m
igr
at
ion
on the
d
e
sti
nation PM
.
Gu
a
ngyo
ng
et
.
al
.
[22
]
,
propo
sed
two
te
ch
ni
qu
e
s
for
i
m
pr
ovin
g
pr
e
-
c
op
y
appr
oach
:
1)
te
chn
i
qu
e
of
m
e
m
or
y
com
p
act
ion
based
on
dis
k
cac
he
a
nd
m
e
m
or
y
sna
psho
t;
2)
sc
hem
e
of
ada
pti
ve
do
wn
ti
m
e
con
t
ro
l
base
d
on
the
hi
story
of
VM’
s
m
e
m
or
y
upda
te
inform
at
ion
cal
le
d
W
rita
bl
e
W
or
king
Set
-
WW
S.
They
hav
e
i
m
ple
m
ented
the
m
et
ho
d
in
KV
M
hype
rv
is
or.
It
s
hows
th
at
the
m
e
m
or
y
com
pacti
on
te
chn
i
qu
e
s
can
r
edu
c
e
m
e
m
or
y
trans
f
er
ti
m
e
by
a
f
act
or
of
2
a
nd
m
os
tl
y
in
the
first
phase.
E
xp
e
rim
ental
resu
lt
s
p
r
ov
e
d
t
hat
the
adap
ti
ve
VM
dow
nti
m
e
cont
ro
l
te
ch
nique
su
ccess
f
ully
handled
the
li
ve
m
igrati
on
for
the
VM
r
unning
m
e
m
or
y
intens
ive
w
orkloa
ds.
But
w
hen
t
he
m
e
m
or
y
con
te
nts
of
ta
r
get
m
igrati
ng
VM
c
hange
t
oo
m
uch
,
the
m
e
m
or
y com
pacti
on
tec
hn
i
que m
igh
t n
ot
work
well
.
3.1.2.
Post
-
c
opy
appro
ach
In
t
his
a
ppro
ac
h
[
23
]
,
eac
h
m
e
m
or
y
pa
ge
ha
s
bee
n
tra
nsfer
red
once
a
nd
this
m
akes
it
bette
r
than
t
he
pre
-
co
py
ap
pro
ach.
T
he
total
m
igrati
on
ti
m
e
and
nu
m
ber
of
trans
ferre
d
pa
ges
are
le
ss
[
2
4
]
.
At
the
be
gi
nn
in
g,
po
st
-
co
py
s
usp
ends
the
m
igra
ti
ng
VM
at
the
source
PM,
it
cop
ie
s
m
ini
m
a
l
process
or
sta
te
to
the
destin
at
ion
PM,
resu
m
es
t
he
VM,
an
d
sta
rts
fetchin
g
m
e
m
or
y
pag
es
f
r
om
the
so
urce
PM
throu
gh
th
e
network.
The
way
how
pa
ges
a
re
fetche
d,
hel
ps
rising
dif
fer
e
nt
var
ia
nts
of
post
-
co
py,
w
he
re
each
is
consi
de
red
as
a
f
unct
ion
f
or
i
m
pr
ovem
ents.
a.
Dem
and
pa
ging:
w
hen
sta
rts
the
m
igrated
V
M
on
destinat
ion
PM,
an
d
th
e
pag
e
s
it
needs
are
not
in
th
e
m
e
m
or
y,
then
pa
ge
fa
ult
oc
cur
s
.
It
ca
n
be
se
r
viced
by
requesti
ng
t
he
pa
ge
th
r
ough
the
n
et
wor
k.
The pa
ge
is t
ra
ns
fe
rr
e
d
a
nd
due to
the t
raffic
V
irt
ual Mac
hi
ne
sl
ow
s
do
wn.
b.
Act
iv
e
pa
gi
ng
:
wh
il
e
VM
is
run
ning,
the
pa
ges
are
pr
oac
ti
vely
pu
sh
e
d
into
it
[25
]
.
If
any
pag
e
fau
lt
s,
tho
se
can
b
e
se
rv
ic
e
d by dem
and p
a
ging a
nd
for
t
ran
s
ferre
d pa
ges n
o p
age
f
a
ults occ
ur.
c.
Pre
-
paging
:
th
is
var
ia
nt
is
li
ke
activ
e
pagi
ng
but
pre
dicti
ng
the
s
pecial
local
it
y
of
VM
m
e
m
or
y
acce
ss
patte
rn
[
23
]
.
T
he next
pa
ges wh
ic
h wil
l be a
ccesse
d
a
re tr
a
ns
fe
rr
e
d
t
o
the
VM.
d.
Dyna
mic
Self
-
Ballo
on
i
ng
(
D
SB)
:
is
us
ed
to
avo
i
d
tran
sf
e
r
of
f
ree
m
e
m
or
y
pag
es.
It
release
s
per
i
od
ic
al
ly
fr
ee
pa
ges
of
VM
back
to
the
hypervis
or
s
,
so
this
w
ay
sp
eeds
-
up
the
process
of
m
igrati
on
with
neg
li
gi
ble
pe
r
form
ance
deg
r
adati
on.
Th
us,
releasi
ng
pro
cess
of
t
hese
pag
es
t
hat
are
not
us
e
d
is
increase
d
to
th
e
destinat
io
n P
M, as a
res
ult o
f
this t
he
total
m
igrati
on
ti
m
e
is re
du
ce
d.
3.1.3.
Hybri
d
techniq
ue
(
Pr
e a
n
d
p
ost c
opy
)
It
us
es
t
he
be
nef
it
s
of
pr
e
-
c
op
y
a
nd
post
-
cop
y
a
ppr
oaches,
an
d
th
e
c
om
bin
at
ion
of
these
tw
o
appr
oach
es
r
e
du
ce
s
serv
ic
e
dow
nti
m
e
and
the
total
m
igrati
on
tim
e.
First,
it
wo
rk
s
as
a
pr
e
-
c
op
y
ap
proac
h
wh
il
e
VM
is
r
unning
on
t
he
so
urce
PM.
A
f
te
r
the
first
it
erati
on
of
VM
m
e
m
or
y
trans
f
er
is
st
oppe
d,
and
it
resu
m
es
at
destinat
ion
PM
wi
th
it
s
pr
ocess
or
sta
te
and
dirty
pag
es.
T
he
n,
the
rem
ai
nin
g
pa
ges
are
tra
ns
f
err
e
d
by post
-
c
opy a
ppr
oach [
2
4
].
Chou
dh
a
ry
et
.a
l.
[21
]
,
com
par
es
pr
e
-
co
py
with
post
-
c
op
y
,
seen
that
the
seco
nd
ap
proa
ch
reduces
th
e
total
m
igrati
on
tim
e
and
the
trans
ferre
d
num
ber
of
pa
ges.
But
du
e
to
m
igrati
on
la
te
ncy
of
fetc
hing
the
pag
es
it
has
m
or
e
do
wn
ti
m
e
than
th
e
first
a
ppr
oac
h.
A
no
t
her
dis
adv
a
ntage
of
post
-
c
op
y
is
that
if
any
fail
ur
e
occurs
durin
g
t
he
m
i
gr
at
io
n,
the
r
ecov
e
ry
m
ay
no
t
be
possib
le
.
Wh
en
pr
e
-
cop
y
or
po
st
-
cop
y
im
pr
ove
s
the
perform
ance,
de
pends
upon
the
w
orklo
ad
t
ype
an
d
pe
rform
ance
go
al
of
m
igrati
on
.
It
is
con
cl
uded
t
ha
t
pr
e
-
cop
y
co
uld
be
consi
der
e
d
as
a
bette
r
appr
oa
ch
f
or
rea
d
int
ensive
work
l
oa
d,
an
d
post
-
co
py
for
w
rite
intensive
or large
m
e
m
o
ry wor
klo
a
d.
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
An overvi
ew
of virt
ua
l
m
ach
i
ne
li
ve migr
atio
n
te
ch
niques
(
Artan M
az
rek
aj
)
4437
Hines
et
.
al.
[2
3
]
,
al
s
o
c
om
par
ed
post
-
cop
y
a
nd
pre
-
co
py
ap
proa
ch
on
Xe
n.
They
sho
w
i
m
pr
ovem
ents
in
so
m
e
m
igrati
on
m
et
rics
lik
e
pa
ges
tra
nsfer
red,
netw
ork
over
hea
d
an
d
total
m
igrati
on
ti
m
e
us
in
g
a
range
of
VM
w
orkl
oa
ds
.
T
hey
m
iti
gate
us
in
g
of
po
st
-
co
py
with
adap
ti
ve
pre
-
pa
gi
ng
to
el
im
inate
duplica
te
of
al
l
pag
e
tra
ns
m
is
sion
s
.
Ying
wei
et
.a
l.
[26
]
,
des
cribes
a
schem
e
fo
r
whole
-
sy
stem
l
ive
m
igrati
on
.
Fo
r
achie
ving
an
inc
onside
ra
ble
dow
ntim
e
and
finite
depend
e
ncy
f
r
om
so
urce
PM
,
thi
s
kind
of
m
igr
at
ion
trans
fer
s
the
w
ho
le
syst
em
run
-
ti
m
e
state,
lik
e
m
e
m
or
y
data,
local
disk
s
tora
ge
an
d
CP
U
sta
te
of
th
e
VM
.
It
proposes
a
TPM
–
Th
ree
-
Ph
ase
Mi
grat
ion
al
gorithm
,
wh
ic
h
com
po
s
ed
of
pr
e
-
co
py
,
fre
eze
-
and
-
c
opy
an
d
po
st
-
co
py
.
3.2.
Fil
e migr
at
i
on
A
co
ns
ist
ent
vi
ew
an
d
a
locat
ion
in
de
pen
de
nt
view
of
file
syst
e
m
sh
ou
ld
be
avail
able
on
al
l
PMs
to
su
pp
or
t
VM
m
igrati
on.
A
s
ol
ution
to
sup
port
this,
is
pro
vi
ding
eac
h
VM
with
it
s
own
virtu
al
disk,
w
hich
is
m
app
ed
in
t
he
file
syst
e
m
,
an
d
it
transpor
ts
t
his
virtu
al
dis
k’s
co
ntents
al
ong
with
the
ot
he
r
sta
te
s
of
the
VM.
Anothe
r
pro
posed
way
co
uld
be
to
ha
ve
a
gl
ob
al
file
syst
em
through
al
l
PMs,
w
he
re
c
ould
be
lo
cat
ed
a
VM,
bu
t i
t i
s
not so
pr
act
ic
al
to p
rovide
a
consist
e
nt g
l
ob
al
r
oo
t
f
il
e syst
e
m
thr
ough all
PMs
.
In
[
27
]
is
ta
ck
le
d
a
distrib
ut
ed
st
or
a
ge
te
c
hnology
for
V
M
m
igrati
on,
wh
ic
h
is
know
n
as
I
nter
net
Su
s
pend/Re
s
um
e
(I
SR).
It
le
ts
su
sp
e
nd
on
on
e
PM
a
nd
se
a
m
le
ssly
resum
e
on
ano
t
her
PM.
A
distrib
ut
ed
file
syst
e
m
us
ed
in
IS
R
ser
ves
to
trans
fer
the
files
of
s
uspen
de
d
VM
sta
te
.
Hyperviso
r
use
s
local
f
il
es
to
store
t
he
con
te
nts
of
ea
ch
VM’s
vi
rtu
al
disk
an
d
th
en
tra
nsfers
th
os
e
a
nd
oth
e
r
inf
or
m
at
ion
of
VM
sta
te
to
the
destinat
io
n.
Th
ere
are
know
n
two
dif
fe
ren
t
te
chn
i
qu
e
s,
one
cal
le
d
sma
rt
cop
yi
ng
a
nd
othe
r
one
is
a
pr
oact
iv
e
state
tran
sfe
r,
wh
ic
h
help
s
to
reduce
the
am
ount
of
data
th
at
hav
e
to
be
trans
ferre
d
fro
m
su
sp
en
de
d
VM
to
resu
m
e V
M.
Also
,
a
nothe
r
s
yst
e
m
wh
ic
h
ta
ckles
the
file
m
igrati
on
is
Z
ap
[26
]
that
pr
ov
i
des
a
thin
vi
rtuali
zat
ion
la
ye
r
that
intr
oduce
s
a
pr
oc
es
s
dom
ain
abst
r
act
ion
known
as
pod
a
nd
is
s
et
on
t
op
of
th
e
OS.
It
pro
vid
e
s
a
set
of
processes
w
it
h
the
sa
m
e
vi
rtuali
zed
vie
w
of
the
syst
em
a
nd
is
in
dicat
ed
by
a
virtu
al
iz
ed
file
syst
e
m
p
rivat
e
and
c
orres
pondin
g
pri
vate
file
syst
e
m
na
m
e
sp
ace
that
pres
ents
this
gro
up.
W
he
n
pods
a
re
create
d
or
th
ey
are
m
ov
ed
to
a
P
M,
al
so
is
crea
te
d
a
pod
ide
nt
ifie
r
on
PM
f
or
each
pod
t
o
se
rv
e
as
a
n
area
for
virt
ual
file
syst
e
m
of
pod
.
Z
ap
ta
ke
s care t
o
m
ake th
e
directo
ry
inacce
ssible
for pr
ocesses
on t
he
PM
w
hich are
no
t i
n give
n pod.
3.3.
Ne
twor
k
mi
gration
Fo
r
rem
ote
com
m
un
ic
at
ion
of
syst
e
m
s
with
VM,
each
V
M
sh
ould
ha
ve
it
s
MAC
add
ress
a
nd
the
virtu
al
I
P
ad
dr
ess.
Hy
perviso
rs
pe
rfor
m
m
a
pp
i
ng
of
virt
ua
l
IP
an
d
the
M
AC
ad
dr
es
ses
to
their
c
orres
pondin
g
VMs. A
ls
o,
is
pro
vid
e
d
an unsoli
ci
te
d
ARP
rep
ly
f
r
om
m
igrati
ng
PM
wh
i
ch
ad
ver
ti
ses t
he
IP
m
ov
e
d
to
a n
ew
locat
ion
i
f
the
m
achines
incl
uded
in
VM
m
i
gr
at
io
n
a
re
c
onnected
with
sw
it
ched
netw
ork
.
S
o,
fu
t
ur
e
pa
ckets
are
se
nt
to
the
ne
w
loc
at
ion
by
rec
onfi
guring
al
l
the
pee
r
s.
T
he
m
igrati
ng
ope
rati
ng
s
yst
e
m
has
an
act
ual
MAC
ad
dr
ess
t
o
detect
it
s
m
o
ve
to
a
new
po
rt
[2
8
]
.
T
her
e
i
s
a
syst
e
m
fo
r
netw
ork
m
igrati
on
,
wh
ic
h
is
known
as
Q
uasar
[
29
]
,
pro
vid
es
s
uppo
rt
f
or
m
igr
at
ion
of
c
om
pu
ti
ng
en
vir
on
m
ents.
It
is
e
quip
ped
with
a
virtu
al
netw
ork
m
echan
ism
wh
ic
h
m
akes
m
igrati
on
t
rans
pa
ren
t
to
guest
OS
e
s.
Th
e
m
echan
ism
allow
s
ne
twork
connecti
ons t
o be
kep
t ac
r
os
s
m
igrati
on
.
3.4.
De
vice m
igra
tion
Nowa
days,
vi
rtuali
zat
ion
is
no
t
su
pport
ed
in
t
he
m
os
t
of
the
hard
war
e,
s
o
de
vice
virtua
li
zat
ion
co
uld
on
ly
rely
on
pure
s
of
twa
re
te
chnolo
gy
[30
].
Ph
ysi
cal
reso
urces
are
sh
a
red
us
ing
s
of
t
ware
based
vi
rtual
iz
at
io
n
betwee
n
dif
fere
nt
gu
e
sts,
by
pr
e
ve
nting
acc
ess
to
dev
ic
e
r
eso
ur
ces
f
ro
m
gu
e
sts.
De
vice
m
igrati
on
nee
ds
tha
t
VMs
ca
n’
t
us
e
PM
s
pecific
de
vices
as
so
m
e
de
vices
are
di
ff
ic
ult
to
m
igrate.
Ge
ner
al
ly
,
there
are
t
hr
ee
ty
pes
of d
e
vice s
upport
pro
vid
e
d
to
m
ake d
evice
m
igrati
on
poss
ible [
2
8
]:
a.
Em
ula
ti
on
,
it
serv
e
s
f
or
em
ulati
ng
a
dev
i
ce
in
s
of
t
war
e
.
F
or
exam
ple,
a
virtu
al
c
onso
le
co
uld
be
reg
ist
ere
d
as
/d
ev/co
nsole
.
b.
Virt
ua
li
za
ti
on
,
if
a
VM
m
igra
te
s
to
a
nother
P
M
wh
ic
h
has
a
n
e
qu
i
valent
de
vice,
the
VM
will
be
a
ble
to
util
iz
e it
.
c.
A
No
n
-
migrata
ble
device
dr
iv
er
,
pa
sses
al
l
r
equ
e
sts
to
t
he
dev
ic
e
on
the
PM,
but
wh
e
n
the
de
vice
is
i
n
us
e it
do
es
not
al
low
m
igrati
on
.
4.
OTHE
R
LI
V
E V
M MIG
R
ATIO
N
TE
C
HNIQ
UES
4.1.
A
daptiv
e
memor
y
c
ompr
ession
Fo
r
bala
ncin
g
the
cost
a
nd
th
e
perform
ance
of
VM
m
igratio
n,
aut
hors
at
[31
]
intr
oduce
an
ad
aptiv
e
zero
-
awa
re
c
om
pr
ession
al
gorithm
.
They
pro
po
se
d
t
he
VM
m
igrati
on
te
chn
i
qu
e
cal
le
d
MEC
OM
that
us
e
s
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.
9
, N
o.
5
,
Oct
ober
201
9
:
4
4
3
3
-
4
4
4
0
4438
m
e
m
or
y
com
p
ressio
n
in
orde
r
to
fast,
a
nd
s
ta
ble
VM
m
igrati
on.
F
ro
m
exp
e
rim
ents
that
com
par
ed
to
Xen,
their syst
em
ca
n red
uce
32%
of total
m
igratio
n t
i
m
e, 2
7.1
%
of
down
ti
m
e and 68.
8%
of to
ta
l t
ran
sfe
rr
e
d data.
4.2.
Usin
g sh
ar
ed stor
ag
e
In
orde
r
t
o
ke
ep
the
dow
ntim
e
to
m
ini
m
u
m
the
aut
hors
at
[
32
]
pr
e
se
nts
a
te
c
hn
i
que
w
hich
will
reduce
the
tota
l
tim
e
to
m
igr
at
e
a
run
ning
VM
f
ro
m
on
e
PM
to
an
ot
her
.
Their
te
c
hn
i
que
m
akes
an
update
d
m
app
in
g
of
m
e
m
or
y
pa
ges
that
presentl
y
exist
in
the
sam
e
for
m
on
the
stora
ge
de
vice
and
trac
ks
t
he
VM’
s
I/O
op
e
rati
on
to
the
netw
ork
at
ta
ched
sto
ra
ge
dev
ic
e
.
T
hro
ugh
the
it
erati
ve
pre
-
co
py
phase,
t
he
m
e
m
or
y
to
disk
m
app
in
g
will
be
sent
to
the
destinat
io
n
PM,
after
t
hat
it
will
get
the
con
te
nts
f
ro
m
the
netw
ork
at
t
a
ched
stora
ge
dev
ic
e,
rathe
r
tha
n
t
ra
ns
fe
rr
i
ng
pa
ge
s
from
the
sou
rce
to
t
he
destinat
ion.
F
r
om
t
he
r
esults
it
is
seen
that a re
duct
io
n of u
p o
ver
30
% in t
otal t
ran
s
fer
ti
m
e fo
r a
ra
ng
e
of
be
nchm
ark
s.
4.3.
Ex
ploitin
g
d
ata de
-
dup
li
cat
ion
Zha
ng
et
.
al
[
33
]
des
c
ribes
a
VM
m
igrati
on
ap
proac
h,
Mi
gr
at
io
n
with
D
at
a
De
-
duplica
ti
on
(MD
D),
wh
ic
h
presents
data
de
-
duplica
ti
on
into
the
m
igrati
on
.
MD
D
has
ha
sh
bas
ed
fin
gerp
rints
to
discover
id
entic
al
m
e
m
or
y
pag
es
,
an
d
in
or
der
to
exclu
de
r
edun
dan
t
m
e
m
or
y
data
when
m
igrati
o
n
and
util
iz
es
the
sel
f
-
si
m
il
arity
of
run
tim
e
m
e
m
or
y
i
m
age
us
es
Run
Le
ng
t
h
E
ncode
(RLE
).
Fr
om
the
exp
e
rim
ents,
i
t
is
s
een
that
MDD
ha
s
reduced
34.93%
of
total
m
igrati
on
tim
e,
56
.
60%
of
total
data
transf
er
re
d
durin
g
m
igratio
n,
an
d
26.16%
of
dow
nti
m
e.
4.4.
Ener
gy aw
are v
ir
tual
machine
migr
at
i
on
The
pa
per
[34
]
,
intro
duces
a
n
al
gorithm
th
at
deal
with
the
load
of
the
PM
and
wh
ic
h
eff
ic
ie
ntly
m
igrates
the
V
Ms.
T
her
e
are
oth
e
r
im
po
rtan
t
facto
rs
t
hat
a
re
us
e
d
to
sel
e
ct
VMs
w
hich
will
be
m
igrated,
s
o
the
ene
rg
y
c
onsu
m
ption
will
be
m
ini
m
iz
e
d
by
s
hutt
ing
dow
n
un
deru
ti
li
zed
PMs.
This
will
cau
se
the
reducti
on
of
th
e
ene
rg
y
c
os
t.
The
a
utho
rs
e
va
luate
their
pr
opos
e
d
s
olu
ti
on
wh
il
e
us
in
g
their
own
sim
ulator
.
Af
te
r
t
he
ex
perim
ents
their
resu
lt
s
show
that
propose
d
m
eth
od
re
duces
en
erg
y
co
nsum
ption
up
to
20.8
%
for
sta
ti
c V
M l
oad an
d up to
22.
0 %
of
dynam
ic
V
M l
oa
d
c
om
par
ed
to p
ur
e
pe
rfor
m
ance b
ase
d VM m
igrati
on
.
4.5.
C
ontinu
al
migration
Cui
et
.
al
[3
5
]
r
epr
ese
nt
that
by
con
ti
nuously
prop
a
gatin
g
sta
te
of
the
VM’
s
to
a
back
up
PM
throu
gh
li
ve
m
igrati
on
te
chn
iq
ues
,
ap
plica
ti
on
s
in
the
VM
with
m
ini
m
al
do
wntim
e
can
be
re
paire
d
f
ro
m
har
dwa
r
e
fail
ur
es
.
F
ro
m
the
res
ults
it
is
seen
t
han
i
n
a
co
ntinu
al
m
igrati
on
syst
e
m
,
if
a
fail
ure
is
de
te
ct
ed,
the
V
M
can
be repaire
d
i
n
l
ess tha
n 1
sec
, a
lt
ho
ug
h per
form
ance i
m
pact
to
the
pr
otect
ed VM ca
n be redu
ce
d
t
o 30%
.
4.6.
As
yn
c
hronous
repli
ca
tion
and
s
tate s
yn
chr
on
iz
at
i
on
Liu
et
.al
[
36
]
exp
la
in
s
that
e
xecu
ti
on
trace
is
logged
on
t
he
s
ource
PM,
in
orde
r
to
c
oor
din
at
e
the
run
ning
source
and
de
s
ti
natio
n
VMs
to
achi
eve
co
ns
ist
ent
sta
te
,
wh
ere
is
us
ed
a
sync
hron
iz
at
io
n
te
ch
nique.
The
a
utho
rs
i
ntr
oduce
t
he
i
m
ple
m
entat
io
n
of
a
new
appr
oach,
CR
/TR
-
Moti
on,
w
hich
a
dopts
c
heck
-
po
i
nting
/
recov
ery
an
d
trace/
r
eplay
te
chnolo
gies
to
pro
vid
e
fast,
tra
nspare
nt
V
M
li
ve
m
i
gr
at
io
n
for
W
AN
a
nd
LAN.
T
he
co
ns
um
ption
of
the
netw
ork
ba
ndwidt
h
an
d
m
igrati
on
dow
nti
m
e
can
be
reduce
d
by
C
R/
TR
-
Moti
on.
Ex
pe
rim
ents
sh
ow
that
the
appro
ac
h
can
re
duce
m
igrati
on
com
par
ed
to
m
e
m
or
y
-
to
-
m
e
m
or
y
te
chn
iq
ue
in
a
LAN,
up
to
31
.5
%
on
t
otal
m
igrati
on
ti
m
e,
up
to
72.
4%
on
a
pp
li
cat
io
n
noti
ced
dow
ntim
e,
an
d
up
t
o
95.
9%
on
the
data
to
s
ynch
ronize
the
VM
sta
te
.
For
a
var
ie
ty
of
w
orkloa
ds
m
igrated
acr
os
s
WANs,
the
app
li
cat
io
n per
form
ance o
ve
r
head f
or m
igrati
on
is
8.5
4 per
cent on a
ve
rage.
4.7.
G
an
g
mi
gratio
n
Live
m
igrati
on
of
m
ulti
ple
VMs
at
the
sa
m
e
tim
e
fr
om
one
group
of
PM
s
to
an
oth
e
r
in
reacti
on
t
o
even
ts
s
uch
as
un
a
vo
i
dab
le
f
ai
lures
load
s
pik
es
is
know
n
as
Gang
Mi
gr
at
io
n
(
GM).
A
la
rg
e
am
o
un
t
of
netw
ork
traf
fi
c
gen
e
rated
by
Gang
m
igrati
on
caus
e
s
ov
erloa
d
on
co
r
e
netw
ork
li
nks
an
d
s
witc
he
s
in
a
datace
nter
.
Usi
ng
gl
ob
al
de
-
duplica
ti
on,
t
he
auth
ors
i
n
[
37]
prese
nt
a
te
chn
i
qu
e
to
kee
p
dow
n
t
he
ne
twork
ov
e
r
head
of
G
M.
The
GM
de
te
rm
ines
an
d
exclu
des
t
he
r
et
ran
sm
issi
on
of
du
plica
te
m
e
m
or
y
pa
ges
be
twee
n
VMs r
un
ning on
num
ero
us
P
Ms i
n
the cluster. Th
e im
plem
entat
ion
o
f G
M glob
al
d
e
-
duplica
ti
on and ev
al
uate
it
us
ing
QEM
U/K
VM
VMs
. F
ro
m
the
res
ults
it
is
seen
that
it
reduces
the
total
m
igrati
on
tim
e
of
VMs
a
nd
th
e
netw
ork
tra
ff
ic
. A
s
w
el
l,
h
as
a
sm
aller r
eve
rs
e p
e
rfor
m
ance
i
m
pact o
n net
work
-
bound ap
plica
ti
on
s.
In
the
Table
1
is give
n
a su
m
m
ary of
sev
e
ra
l t
echn
iq
ues
f
or V
M Li
ve
m
igrati
on. In
this s
umm
ary are
consi
der
e
d
the
m
os
t
i
m
po
rtant
featu
res
on
li
ve
m
igrati
on
,
su
ch
as:
ba
sic
m
igrati
on
te
chn
i
qu
e
,
nam
e
of
t
he
te
chn
iq
ue, hy
pe
rv
is
or u
se
d, m
et
rics, and a
chi
evem
ents/bene
fits.
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
An overvi
ew
of virt
ua
l
m
ach
i
ne
li
ve migr
atio
n
te
ch
niques
(
Artan M
az
rek
aj
)
4439
Table
1.
C
om
par
iso
n of VM
l
ive m
igrati
on
t
echn
i
qu
e
s
Bas
ic
m
ig
ration
tech
n
iq
u
e
Na
m
e
of
the
tech
n
iq
u
e
Hy
p
ervis
o
r
u
sed
Metr
ics
Ach
iev
e
m
en
ts/B
en
ef
its
Ref
Pre
-
co
p
y
Alg
o
rith
m
Sh
ared
sto
rage
Xen
-
Migr
atio
n
ti
m
e
-
Do
wn
ti
m
e
-
Net
wo
rk traff
ic
-
Nu
m
b
e
r
o
f
pag
es
trans
f
err
ed
3
2
%
to
tal
m
ig
r
ati
o
n
ti
m
e;
Nu
m
b
e
r
o
f
p
ag
es trans
f
err
ed
i
s al
m
o
st eq
u
al;
3
7
% o
f
d
o
wn
ti
m
e
.
[
1
9
,
2
4
,
3
2
]
Ad
ap
tiv
e
Me
m
o
r
y
Co
m
p
r
ess
io
n
Xen
3
2
%
to
tal
m
ig
ration
ti
m
e
;
6
8
.8%
n
u
m
b
er
o
f
p
ag
es
trans
f
err
e
d
redu
ced;
2
7
.1%
do
wn
ti
m
e.
[
3
1
]
Po
st
-
co
p
y
Alg
o
rith
m
Ad
ap
tiv
e pre
-
p
ag
in
g
Xen
-
Nu
m
b
er
o
f
pag
es
trans
f
err
ed
Eli
m
in
ate
all
d
u
p
licate
p
ag
e
trans
m
iss
io
n
s;
Red
u
ce
th
e
n
u
m
b
e
r
o
f
n
etwo
rk
-
b
o
u
n
d
p
ag
e f
au
lts to
within
21
%.
[
2
3
,
2
5
,
14]
[
3
3
]
Pre
-
co
p
y
alg
o
rith
m
Exp
lo
itin
g
Data
De
d
u
p
licatio
n
Xen
-
Migr
atio
n
ti
m
e
-
Do
wn
ti
m
e
Red
u
ces
5
6
.60
%
o
f
to
tal
d
at
a
trans
f
err
ed
du
ring
m
ig
ration
;
3
4
.93
% o
f
total
m
i
g
ration
ti
m
e.
Live
m
ig
r
atio
n
Pre
-
co
p
y
Energy
awar
e
VM
alg
o
rith
m
Cu
sto
m
bu
ilt
si
m
u
l
ato
r
si
m
u
l
ato
r
Po
wer
co
n
su
m
p
t
io
n
Red
u
ces
Po
wer
Cons
u
m
p
tio
n
,
2
8
%
o
n
static lo
ad
and
22
% o
n
dyna
m
ic
lo
ad
.
[
3
4
]
Live
m
ig
r
atio
n
Pre
-
co
p
y
Co
n
tin
u
al
m
ig
ration
KVM
Fau
lt toleran
ce
Hardwar
e
f
ailu
re
(
recov
ered
less
th
an
o
n
e sec)
;
Red
u
ced Do
wn
ti
m
e,
3
0
% o
v
erall
p
erfo
rm
a
n
ce.
[
3
5
]
Live
m
ig
r
atio
n
p
re
-
co
p
y
Asynch
ron
o
u
s
replicatio
n
and
state
synch
ron
izatio
n
UML
in
u
x
/
Re
-
Virt
-
lo
g
an
d
r
ep
lay
to
o
l
Do
wn
ti
m
e
Migratio
n
ti
m
e
Netwo
rk traff
ic f
au
lt
Tolerance
Red
u
ces 7
2
,4%
downti
m
e;
Red
u
ces 3
1
,4%
m
i
g
ration
ti
m
e;
9
5
.9%
netwo
rk
b
an
d
wid
th
.
[
3
6
]
Live
m
ig
r
atio
n
Gan
g
m
ig
ration
QEM
U/KV
M
Migratio
n
ti
m
e
Netwo
rk b
an
d
wid
th
Red
u
ces 4
2
%
m
ig
r
atio
n
ti
m
e;
Red
u
ces 6
5
% n
etwo
rk b
an
d
wid
th
5.
CONCL
US
I
O
N
This
pap
e
r
is
an
ov
e
r
view
of
li
ve
m
igrati
on
of
VM
te
c
hniq
ues.
The
li
ve
m
igrati
on
i
nvolv
e
s
the
process
of
m
ov
in
g
VM
or
m
ul
ti
ple
VMs
from
on
e
ph
ysi
cal
m
achine
to
a
no
t
her,
wh
il
e
they
’r
e
run
ning.
Ser
vices
that
are
run
ning
on
VM’s
m
us
t
be
avail
able
to
the
us
e
rs,
s
o
th
ey
will
be
m
igrated
w
hile
they
are
run
ning.
The
r
easo
ns
for
li
ve
VM
m
igrati
on
a
re:
sys
te
m
m
ai
ntenan
ce,
load
balancin
g,
powe
r
m
anage
m
ent
,
proacti
ve fa
ult t
oleran
ce
, reso
ur
ce
sh
a
rin
g.
The
pa
pe
r
f
ocuses
on
t
he
com
pr
e
he
ns
ive
li
te
ratur
e
rev
ie
w
of
oth
e
r
w
ork
re
fer
s
a
nd
try
in
g
to
br
i
ng
t
o
the
resea
rch
e
r
s’
un
der
sta
nd
i
ng
of
li
ve
m
i
gr
at
io
n
te
ch
ni
qu
e
s
by
des
cri
ption
of
wea
knesses
a
nd
st
r
eng
t
hs
,
key
as
pects
of
m
igrati
on
li
ke
CPU,
m
e
m
or
y,
netw
ork
,
a
nd
stora
ge.
The
di
scussion
is
co
nc
entrated
in
som
e
of
perform
ance
m
et
rics,
li
ke:
dow
nti
m
e,
tot
al
m
igrati
on
ti
m
e,
per
f
or
m
ance
deg
ra
datio
n,
et
c.,
that
aff
ect
t
he
process
of
V
M
li
ve
m
igratio
n.
Cl
assifi
cat
ion
of
li
ve
m
igrati
on
te
ch
ni
ques
by
ex
plain
ing
th
ree
basic
on
es:
pre
-
co
py, p
os
t
-
cop
y a
nd
hybri
d
li
ve VM m
igrati
on.
REFERE
NCE
S
[1]
G.
Singh
and
P.
Gupta,
"A
rev
i
ew
on
m
igra
ti
o
n
te
chn
ique
s
an
d
cha
l
le
ng
es
in
li
ve
v
irt
ua
l
m
achine
m
igra
t
ion,
"
2016
5th
Inte
rn
ati
onal
Conf
ere
nce
on
Reliabili
ty
,
In
foc
om
Tec
hnologi
es
and
Optimizati
on
(
T
rends
and
Fut
ure
Direc
ti
ons)
(
ICRITO
)
,
Sep
2016
,
pp.
542
-
546
.
[2]
T.
Erl,
R
.
Putt
ini,
and
Z
.
Mahm
o
od,
Cloud
Computing:
Con
ce
pts,
Technol
og
y
&
A
rchit
e
ct
ure
,
P.
Hall Press,
2013.
[3]
"V
MW
are
,
"vS
p
her
e
ESX
and
ESXi
Info
Cent
er
,
"
[Online
]
,
Available:
htt
ps://
ww
w.vm
ware
.
com
,
[
ac
ce
ss
ed
on
Ma
y
2018
]
.
[4]
Microsoft,
"W
indows
Virtua
l
PC
,
"
[Online
]
,
Available:
htt
p:
//w
ww
.
m
ic
rosoft.
com/windows
/vi
rtua
l
-
pc
,
[
a
ccess
ed
on
June
2018
]
.
[5]
Xen,
"X
en
H
y
p
e
rvisor
,
"
[Onlin
e]
,
Avai
la
b
le
:
htt
p
:
//
ww
w.xe
n.
org
/p
roduc
ts/
xenh
y
p
.
html
,
[
a
ccess
ed on Ma
y
2018
]
.
[6]
Microsoft,
"H
y
per
-
V
Serve
r
2012
,
"
[Online]
,
Avail
able:
ww
w.m
ic
rosoft.
com/serve
r
-
cl
ou
d/h
y
p
er
-
v
-
serve
r
,
[
ac
c
essed
on
Jun
e
2018
]
.
[7]
KV
M,
Kerne
l
-
b
ase
d
Virtu
al Ma
chi
ne
,
"
[Onlin
e]
,
Available:
ww
w.l
inux
-
kvm
.
org
,
[
a
ccess
ed
on
Ju
ne
2018
]
.
[8]
Orac
le,
”Vir
tu
alBox
,
”
[Onl
ine
]
,
Avail
ab
le
:
ww
w.vi
rtu
al
box.
o
rg
,
[
ac
c
essed
on
Ma
y
2018
]
.
[9]
X.
Feng,
J
.
T
an
g,
X.
Luo,
and
Y.
Jin,
“
A
per
f
orm
a
nce
stud
y
of
li
ve
VM
m
igra
ti
on
technolog
ie
s:
VM
oti
on
v
s
XenMotion.
Net
work
Archi
te
ct
u
res,
Mana
gemen
t,
and
Applications
IX,”
Asia
Comm
unic
ati
ons
and
Phot
onic
s
Confe
renc
e
an
d
Ex
hibition
(
ACP
)
,
Nov
2011.
[10]
Q.
Z
hang
,
E.
Gurs
es,
R.
Bout
ab
a,
J.
Xi
ao,
“
D
y
n
amic
Resourc
e
Alloc
a
ti
on
for
S
pot
Marke
ts
in
Clouds
,
”
The
11
th
USENIX
con
f. o
n
Hot
top
ic
s
in m
anageme
nt
of
i
nte
rnet
,
cl
oud,
a
nd
enterpr
ise
ne
t
works and
servi
c
es
,
ACM
,
2011
.
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.
9
, N
o.
5
,
Oct
ober
201
9
:
4
4
3
3
-
4
4
4
0
4440
[11]
W
.
Hu,
A.
Hick
s,
L.
Zh
ang,
E.
M.
Dow
,
V.
Soni,
H.
Jiang
,
R.
B
ull
,
and
J.
N.
Matt
h
ews,
"A
Quanti
tative
Stud
y
o
f
Virtua
l
Mac
hin
e
Li
v
e
Migr
at
ion
,
"
In
ACM
Cloud
and
Aut
onomi
c Com
puti
ng
Conf
ere
nce
,
New
York,
Aug 2013
.
[12]
S.
Akous
h,
R.
Sohan,
A.
Rice,
A.
W
.
Moore,
and
A.
Hopper,
"P
red
ic
ti
ng
th
e
Perform
anc
e
of
Virtua
l
Ma
chi
n
e
Migrat
ion
,
"
IE
E
E
Int
ernati
onal
Symposium
on
Mode
li
ng
,
Ana
lys
is
Simulat
ion
o
f
Computer
and
Tele
communic
a
t
ion
Syste
ms
(
MASCOTS
)
,
pp.
37
–
46
,
2010
.
[13]
G.
Soni
and
M.
Kalr
a
,
"Co
m
par
at
ive
Stud
y
of
L
ive
V
ir
t
ual
Ma
chi
ne
Migrat
ion
Tech
nique
s
in
Clou
d,
"
Int.
J. Comput.
Appl
.
,
vo
l. 84, p
p.
19
-
25
,
De
c
20
13.
[14]
M.
R.
Hines,
U.
Deshpande
,
and K.
Gopala
n,
"P
ost
-
cop
y
L
ive
Mi
gra
ti
on
of
Virtual
Mac
hine
s,"
AC
M
SIGO
PS
Oper
Syst
R
ev
,
vol
.
43
,
pp
.
14
-
26
,
Jul
2
009.
[15]
M.
Fors
m
an
and
A.
Glad
,
"A
u
tomate
d
Li
ve
Migrat
ion
of
Vir
t
ual
Ma
chi
nes,
"
Msc
the
sis,
Ble
ki
nge
Instit
u
te
o
f
Technol
ogy
,
Kar
lskrona,
Sw
ed
en
,
Sep
2013.
[16]
X.
Qin,
W
.
Zhang,
W
.
W
ang
,
J.
W
ei
,
X
.
Zh
ao,
and
T
.
Huang
,
"
Towa
rds
a
Cost
-
Aw
are
Data
Mi
gra
ti
on
Appro
ach
for
Ke
y
-
Valu
e
S
tore
s,"
I
EE
E
Int
e
rnational
Conf
ere
nce on
C
luste
r
Computing
,
Oc
t
2012
,
p
p
.
551
–
5
56.
[17]
U.
Deshpande
,
X.
W
ang
and
K.
Gopala
n,
"Live
Gang
Migrat
ion
of
Virtua
l
Ma
ch
ine
s,"
in
Proce
e
dings
of
the
20
t
h
Inte
rnational
Sy
mpos
ium
on
High
Pe
rform
ance
Distribute
d
,
Ne
w Y
ork,
pp.
135
-
146,
Jun
2011
.
[18]
Y.
Luo,
B
.
Zh
an
g,
X.
W
ang,
Z.
W
ang,
Y.
Sun,
H.
Chen,
"L
ive
and
inc
r
emental
whole
-
s
y
stem
m
igra
ti
on
of
virt
u
al
m
ac
hine
s using
bloc
k
-
bit
m
ap
,
"
I
EE
E
Inte
rnat
ion
al
Conf
ere
nce o
n
Cluste
r Comp
uti
ng
,
Sep
2008
,
pp.
99
–
106
,
.
[19]
C.
Clark,
K
.
F
ra
ser,
S.
Hand
,
J.
G.
Hansen,
E.
Jul,
C
.
L
impach,
I.
Pratt,
and
A.
W
arf
ield,
"
Li
ve
Mi
gra
ti
on
of
Virtu
a
l
Mac
hine
s,"
Pro
ce
ed
ings
of
the
2nd
Confe
renc
e
on
Symposium
o
n
Net
worked
Sy
stems
Design
&
Imple
mentat
ion
,
vol.
2
,
pp
.
273
–
2
86,
Ma
y
2005.
[20]
D.
Kapil,
E
.
S.
P
il
li,
and
R
.
C.
Jos
hi,
“
Li
ve
Virtua
l
Mac
hi
ne
Migration
T
ec
hniqu
es:
Surve
y
and
R
ese
ar
c
h
Chal
le
ng
es
,
” 3r
d
IEEE
In
te
rnati
onal
Ad
vanc
e
C
omputing
Conf
ere
nce (
IACC)
,
2012
,
pp
.
963
-
969
.
[21]
A.
Choudhar
y
,
M.
Chandra
Govil,
G.
Singh
,
L
.
K.
Aw
asthi
,
E
.
S.
Pill
i,
and
D.
Kapil
,
“
A
cri
tic
al
surve
y
of
li
v
e
virt
ual
m
ac
hin
e
m
igra
ti
on
te
ch
nique
s
,
”
Journa
l
of
Cloud
Co
mputing:
Adv
an
ce
s,
S
yste
ms
a
nd
Appl
i
cat
ions
,
vol.
6
(3)
,
2017
.
[22]
G.
Piao,
Y.
Oh,
B.
Sung,
and
C
.
Park,
“
Eff
icient
Pre
-
Cop
y
Li
ve
Migrat
ion
wi
th
Mem
ory
Com
pac
ti
on
and
Adapt
i
ve
VM
Dow
nti
m
e
Control
,
”
I
EEE
Fourth
In
ter
nati
onal
Conf
ere
nce
on
B
ig
Data
and
Cl
oud
Computing
,
Dec
2014
.
[23]
R.
M.
Hines
and
K.
Gopala
n
,
“
Pos
t
-
Cop
y
Based
Li
ve
Vir
tua
l
Ma
chi
ne
Mi
gr
at
ion
Us
ing
Adapti
ve
Pre
-
Paging
an
d
D
y
namic
Sel
f
-
B
al
looni
ng
,
”
Proce
ed
ings
of
the
2009
ACM
SIGPLAN/SIGOPS
i
nte
rnational
con
fe
renc
e
on
Vi
rtu
al
ex
e
cut
ion
en
vi
ro
nments
,
Mar
200
9
,
pp
.
51
-
60
.
[24]
S.
Sharm
a,
and
M.
Chawla
,
“
A
te
chnica
l
rev
ie
w
for
effic
ie
nt
vir
t
ual
m
ac
hine
m
ig
rat
ion
,
”
Inte
rnat
i
onal
Confe
renc
e
on
Cloud
and
Ubiquit
ous Comp
uti
ng
and
Eme
rg
ing
Techno
logi
e
s (
CUBE)
,
Sep
2013.
[25]
D.
Pere
z
-
Bote
ro
,
"A
Brie
f
Tut
ori
al
on
Li
ve
Virtu
al
Mac
hin
e
Migrat
ion
from
a
Secur
ity
Perspe
ct
iv
e
,
"
Univer
sit
y
of
Prince
ton
,
US
A,
2011.
[26]
S.
Os
m
an,
D.
Subhrave
t
i,
G.
Su
,
and
J.
Ni
eh,
“
The
Design
and
Im
ple
m
ent
at
ion
of
Za
p:
A
S
y
s
t
em
for
Migrat
in
g
Com
puti
ng
Environm
ent
s
,
”
Proc
ee
dings
of
th
e
5t
h
Symposium
on
Operating
System
s
Design
and
I
mpleme
ntation
,
Boston,
MA
,
De
c
2002.
[27]
M.
Sat
y
ana
r
a
y
an
an,
B.
Gilbe
r
t,
M.
Toups,
N.
T
oli
a
,
A.
Surie
,
D
.
R.
O’H
allar
on
,
A.
W
olba
ch,
J.
Harke
s,
A.
Perri
g,
and
D.
J.
Farbe
r,
“
Perva
sive
Personal
Com
puti
ng
in
an
Int
ern
et
Sus
pend/Resum
e
Sy
st
em
,
”
IEE
E
Int
erne
t
Computing
,
vo
l.
11
(2), 2007.
[28]
S.
Venka
te
sha
,
S.
Sadhu,
and
S.
Kinta
li,
“
Sur
ve
y
of
Virtu
al
Mac
hine
Migra
t
ion
Te
chn
ique
s,
”
Unive
rs
it
y
o
f
Cali
fornia, Santa Barbara,
Mar
2009.
[29]
K.
Onoue,
Y.
O
y
ama,
A.
Yo
nez
awa
,
“
A
Vi
rtua
l
Mac
h
ine
Migrat
ion
S
y
st
e
m
Based
on
a
CP
U
E
m
ula
tor
,
”
Proce
ed
ings
of
the
2nd
Int
er
nati
onal
Wor
kshop
on
V
irtual
izati
on
Techno
l
ogy
in
Distrib
ute
d
Computin
g
,
Nov 2006.
[30]
E.
Zha
i
,
G.
D.
Cum
m
ings,
and
Y.
Dong,
“
Dong
Li
v
e
Migra
ti
on
with
Pass
-
throu
gh
Device
for
L
inux
VM
2008
,
”
Li
nux
S
ymp.
,
v
o
l.
2
,
p
p
.
261
-
267
,
2008
.
[31]
H.
Jin,
L.
Deng
,
S.
W
u,
X.
Shi,
X.
Pan,
“
Li
ve
virt
ual
m
ac
h
ine
m
igra
ti
on
with
ada
pt
ive
m
emory
compress
ion
,
”
IEE
E
Inte
rnat
io
nal
Conf
ere
nce
on
Cluste
r Com
puti
ng
and
Work
shops
,
Oct
2009
.
[32]
C.
Jo,
E
.
Gus
tafs
son,
J.
Son,
B.
Egge
r
,
“
Effici
en
t
li
v
e
m
igration
of
vir
tua
l
m
ac
hine
s
using
share
d
storag
e
,
”
Proce
ed
ings
of
the
9th
ACM
S
IGPLAN/SIGOPS
int
ernati
ona
l
conf
ere
n
ce
on
Vi
rtual
ex
e
cut
i
o
n
env
ironments
,
vol.
48
(7), pp. 4
1
-
50,
Jul
2013.
[33]
X.
Zha
ng
,
Zh
.
Huo,
J.
Ma,
an
d
D.
Meng,
“
Expl
oit
ing
Data
Dedupli
c
at
ion
t
o
Acc
e
le
ra
te
Live
Virtu
al
Ma
c
hine
Migrat
ion
,
”
IEEE
Int
ernati
onal
Confe
renc
e
on
C
luste
r Computin
g
,
Oct
2010.
[34]
M.
H.
Al
Sh
a
y
eji
and
M.
D
.
Sam
raj
esh,
“
An
Ene
rg
y
-
awa
re
Virtual
Mac
hine
Migra
ti
on
Algorit
hm
,
”
Sec
ond
In
te
rnati
onal
Conf
ere
nce on
Ad
vances i
n
Computing
and
Comm
unic
ati
ons (
ICACC2012)
,
Aug 2012.
[35]
W
.
Cui,
D.
M
a,
T.
W
o,
and
Q.
Li
,
“
Enha
nci
ng
rel
i
ability
for
virt
ual
m
ac
hine
s
via
con
t
inua
l
m
igra
ti
on
,
”
15th
Int
ernati
on
al
Conf
ere
nce o
n
Parallel
and
D
istribut
ed System
,
Dec
2009
,
pp
.
937
-
942
.
[36]
H.
Li
u
,
H.
Jin,
X
.
L
ia
o
,
C.
Yu,
an
d
C.
Z
.
Xu,
“
Li
ve
virt
u
al
m
a
chi
ne
m
igra
ti
on
v
ia
as
y
nchr
onous
rep
l
i
ca
t
ion
and
sta
te
s
y
nchr
oni
zation,”
IE
EE
Tr
ans
act
ions o
n
Paral
le
l
and
Distribute
d
Syste
ms
,
vol
.
22
(12),
2011
.
[37]
U.
Deshpande,
B.
Schl
inke
r
,
E
.
Adler
,
and
K.
Gopala
n,
“
Gang
Migration
of
Vi
rtua
l
Mac
hin
es
using
Cluste
r
-
wi
de
Dedupli
c
at
ion
,
”
13th
IE
EE
/
ACM
Inte
rnat
ional
S
y
mpos
ium
on
Clu
ster,
Cloud
,
and
Gr
id
Computing
,
Ma
y
2013.
Evaluation Warning : The document was created with Spire.PDF for Python.