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
.
6
,
Decem
ber
201
9
, p
p.
5211
~
52
17
IS
S
N:
20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v
9
i
6
.
pp5211
-
52
17
5211
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Novelty
cir
cular
neighbo
rin
g t
ech
niq
ue using re
ac
t
ive fault
toleranc
e metho
d
Ah
m
ad
Shu
kr
i Mohd
N
oo
r
, Nur
Far
hah
Mat
Z
ian, N
oor H
afhiz
ah
A
bd Ra
him,
Rabiei
Mam
at,
W
an
Nu
r
Ami
ra
W
an
Az
man
Depa
rtment
o
f
C
om
pute
r
Scie
n
ce,
School
of
In
for
m
at
ic
s a
nd
Appl
ie
d
Ma
the
m
at
i
cs
,
Univer
sit
i
Mal
a
y
sia
T
ere
ngg
anu
,
Mal
a
y
s
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Dec
20
, 201
8
Re
vised
A
pr 18
, 2
01
9
Accepte
d
J
un
11
, 201
9
The
availa
b
il
i
t
y
of
the
dat
a
in
a
distri
bute
d
sy
stem
ca
n
be
i
ncr
ea
se
b
y
implementi
ng
fa
ult
tolera
n
ce
m
e
cha
nism
in
the
s
y
stem.
R
ea
c
ti
v
e
m
et
hod
in
fau
lt
tolera
nc
e
m
ec
hani
sm
dea
l
s
with
rest
art
ing
the
failed
s
erv
i
ce
s,
p
lacin
g
red
undant
copie
s
of
data
in
m
ul
ti
p
le
nodes
ac
ro
ss
net
work,
in
o
the
r
words
dat
a
rep
l
ic
a
ti
on
and
m
igra
ti
ng
th
e
data
for
rec
ov
er
y
.
Eve
n
if
th
e
ide
a
o
f
da
t
a
rep
lication
is
solid,
the
ch
al
l
enge
is
to
choose
the
right
rep
li
c
ation
te
chni
qu
e
tha
t
ab
le
to
pr
ovide
better
da
ta
avail
abi
l
ity
a
s
well
as
consiste
nc
y
th
a
t
invol
ves
rea
d
and
write
oper
at
ions
on
the
red
undant
cop
i
es.
Circ
ul
ar
Neighbor
ing
Re
pli
c
at
ion
(CNR)
te
chn
ique
exploits
nei
ghbor
ing
poli
c
y
in
rep
licating
the
d
at
a
item
s
in
the
s
y
stem
per
form
s
well
with
reg
ar
ds
to
lower
copi
es
ne
ede
d
to
m
ai
n
t
ai
n
the
s
y
stem
ava
i
la
bi
li
t
y
at
the
h
ighe
st
.
In
a
per
form
an
c
e
anal
y
s
is
with
exi
sting
te
chn
iq
ues,
result
s
sho
w
tha
t
CNR
improves
sy
st
e
m
ava
il
ability
b
y
av
era
g
e
37%
b
y
offe
r
ing
onl
y
two
rep
li
c
as
nee
ded
to
m
ai
nt
ai
n
da
ta
av
ai
l
ability
and
consistenc
y
.
Th
e
stud
y
d
emons
tra
te
s
the
poss
ibi
lit
y
o
f
the
proposed
t
ec
hniqu
e
and
th
e
pote
n
ti
a
l
of
d
epl
o
y
ing
in
la
rge
r
and
compl
ex
env
ironment.
Ke
yw
or
d
s
:
Distrib
uted
c
om
pu
ti
ng
Fault t
olera
nce
Re
plica
ti
on
Syst
e
m
av
ai
la
bili
t
y
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
:
Ah
m
ad
S
hukri
Mohd
Noo
r,
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce,
Scho
ol of
Inform
at
ic
s an
d Appli
ed
Ma
t
hem
atics,
Un
i
ver
sit
i M
al
ay
sia
Tereng
ga
nu,
21030 K
uala T
ereng
ganu, Ter
engga
nu, Mal
a
ysi
a
.
Em
a
il
:
ashu
kri
@u
m
t.edu
.m
y
1.
INTROD
U
CTION
Since
distrib
uted
c
om
pu
ti
ng
com
pr
ise
s
of
physi
cal
and
vir
tual
i
m
ages
in
a
la
rg
e
netw
ork
[
1],
neith
e
r
the
volum
e,
siz
e,
locat
ion,
tim
e
pr
ocessi
ng
a
nd
sto
rage
resou
rces
a
re
detai
le
d
a
nd
rest
rict
ed
[
2
,
3].
The
ch
aracte
ris
ti
cs o
f
distri
bu
t
ed
com
pu
ti
ng rem
ov
es r
est
rict
ion
s in usi
ng
t
he
syst
em
an
d
tradit
ion
al
networks
in
pro
vid
in
g
it
s
serv
ic
es
to
t
he
us
e
r
[
1].
H
ow
e
ve
r,
due
t
o
this,
new
prob
le
m
s,
restrict
ion
s
an
d
c
halle
ng
e
s
m
igh
t
arise
f
or
us
er
s
an
d
a
pp
l
ic
at
ion
s
[
3].
T
her
e
fore,
on
to
p
of
the
se
ad
ve
rsity
,
it
is
i
m
po
rtant
to
sa
fegu
ard
i
ng
the
abili
ty
of
the
com
pu
ti
ng
syst
e
m
in
han
dling
unex
pect
ed
fa
ults
and
f
ai
lures
by
pr
ovidin
g
a
reli
ab
le
fau
lt
tolerance
m
ech
anism
[4
]
.
W
it
h
the
pr
e
sence
of
fau
lt
tole
rance
m
echan
is
m
,
syst
e
m
able
to
detect
,
ide
ntif
y
an
d
recover
from
fail
ur
es
with
out
causin
g
m
ajor
dam
age
to
the
fi
nal
outp
ut
of
the
distrib
uted
c
om
pu
ti
ng
[
5]
.
Hen
ce
,
t
he
distrib
uted
syst
e
m
is
ca
pab
le
of
hav
i
ng
a
n
op
ti
m
u
m
and
acce
ptable
perf
or
m
ance
i
n
the
pr
e
sence
of f
a
ult
[
6].
On
e
of
the
w
ay
to
i
m
ple
ment
fau
lt
toler
ance
in
distri
bute
d
c
om
pu
ti
ng
is
by
ap
plyi
ng
reacti
ve
m
et
ho
d
[
3]
w
he
re
w
he
n
a
fa
ul
t
occurs,
a
ppli
cat
ion
will
restart
at
the
la
st
po
i
nt
on
wards,
rep
li
cat
es
the
data
befor
e
m
igratin
g
t
o
oth
e
r
sim
il
ar
m
achine
[
7]
.
Data
re
plica
ti
on
is
a k
ey
p
ra
ct
ic
es
us
ed
f
or
d
isg
uisin
g
er
r
ors
by
pr
ese
r
ving
sep
arate
cop
ie
s
of
data
or
ob
j
ect
and
sync
hro
ni
zat
ion
of
upda
ti
ng
the
data
in
it
s
rep
li
ca
[6
,
8].
It
will
deal
with
w
hen
an
d
w
her
e
to
c
op
y
the
data,
resour
ce
opti
m
iz
at
ion
and
gr
ow
i
ng
or
s
hri
nkin
g
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.
6
,
Dece
m
ber
201
9
:
5211
-
5217
5212
the
re
plica
ti
on
tree
[6
]
.
T
he
te
chn
iq
ue
al
lows
a
prom
pt
and
l
ocall
y
acce
ssed
to
s
har
e
d
da
ta
al
lowi
ng
the
syst
em
to
con
ti
nue
t
he
a
vaila
bili
ty
of
the
a
pp
li
cat
ion
by
ens
ur
i
ng
al
te
rn
at
e
data
ac
cess
opti
on
is
exist.
Con
se
quently
,
offer
i
ng
a
data
rep
li
cat
ion
m
echan
ism
and
con
t
ro
l
pr
oto
c
ol
that
can
pr
om
ise
data
int
egr
it
y,
easy
acce
ss,
reli
abili
ty
and
avail
abili
ty
is
hig
hly
desire
d
to
achiev
e
a
decen
t
distr
ibu
te
d
e
nv
i
ron
m
ent
especial
ly
if it i
s larg
e i
n
siz
e
or h
ea
vy in
tra
ns
act
io
n nu
m
ber
[4].
The
rest
of
t
he
pa
per
is
orga
ni
zed
as
fo
ll
ows
:
Sect
ion
2
dis
cusses
seve
ral
rep
li
cat
ion
te
chn
i
qu
e
s
a
nd
pr
ese
nts
the
re
search
backg
round
of
the
ne
w
ci
rcu
la
r
neig
hbori
ng
te
ch
nique.
Sect
io
n
3
de
scribes
the
ci
r
cular
neig
hbori
ng
m
od
el
with
the
possible
m
ai
nte
nan
ce
protoc
ol
us
ing
reacti
ve
m
et
ho
d.
Sect
ion
4
pr
e
sents
a
ro
ug
h
evaluati
on
on t
he new
tech
nique. C
oncl
us
io
n an
d fu
t
ur
e
w
orks
a
re
giv
e
n
in
Sect
ion
s
5.
2.
R
EPL
IC
ATI
ON
TE
CHNI
QUES
Data
rep
li
cat
ion
c
om
pr
ise
s
of
tw
o
m
a
i
n
ty
pes
of
s
olu
ti
on;
synch
ron
ou
s
a
nd
a
synch
ron
ou
s
rep
li
cat
ion
[9
]
.
Async
hrono
us
rep
li
cat
ion
im
m
ediat
el
y
capt
ur
es
t
he
cha
nges
on
th
e
pr
im
ary
rep
li
ca
an
d
then
tim
e
ly
pr
opag
at
e
the
cha
ng
e
s
[9
]
.
T
his
re
s
ulted
in
a
l
ower
co
st,
flexi
bl
e
and
scal
a
ble
so
luti
on
al
th
ough
deali
ng
with
the
possi
bili
ty
of
data
lost
an
d
net
work
ba
ndwi
dth
[
9].
T
hi
s
ty
pe
of
so
l
ut
ion
w
ork
rea
s
on
a
bly
well
f
or
a
sin
gl
e
obj
ect
update
s
but
un
reli
able
w
hen
in
vo
l
vin
g
m
ulti
ples
ob
j
ect
s
.
F
or
syn
chro
nous
re
plica
ti
on,
it
deals
with
quor
um
to
upda
te
the
re
plica
a
t
the
sam
e
time
an
d
will
ro
ll
back
if
on
e
fai
ls
[10].
Sy
nchr
onous
rep
li
cat
ion
bre
aks
r
eplic
a
int
o
se
ver
al
sc
he
m
es
prov
i
ding
the
hi
gh
avail
abili
ty
,
auto
fa
il
-
ov
e
r
a
nd
m
i
nim
a
l
data
loss
[9
]
.
Howe
ver,
network
e
ff
ic
ie
nc
y,
scal
abili
ty
,
cost
and
le
ss
fl
exibili
ty
issue
cou
l
d
hinde
r
the
abili
ty
of
t
his
te
ch
nique
[
1].
I
n
t
his
pap
e
r,
we
will
fo
c
us
on
sync
hro
nous
re
plica
ti
on
as
t
he
s
olu
t
ion
offer
s
su
it
a
bili
t
y
in
av
oid
in
g
a
nd
res
olv
i
ng
c
onflic
ts
in
re
plica
acce
ss
an
d
up
dates
w
hic
h
prov
i
des
a
m
echan
ism
that
able
to
increase t
he da
ta
av
ai
la
bili
ty
.
2.1.
Re
ad
-
on
c
e
-
w
ri
te
-
all
(ROWA
)
This
is
the
m
os
t
com
m
on
and
strai
ghtf
orwa
rd
po
li
cy
[6
]
use
d
in
re
plica
ti
ng
the
syst
em
wh
ic
h
kee
ps
m
ul
ti
ple
cop
ie
s
of
re
plica
s
th
at
al
low
a
nyon
e
can
be
rea
d
and
m
us
t
al
l
be
update
d.
I
n
t
his
te
ch
nique,
a
rea
d
op
e
rati
on
nee
ds
on
ly
one
c
opy
an
d
is
al
lowed
t
o
rea
d
a
ny
co
py
of
t
he
data.
At
the
m
eantim
e,
a
wr
it
e
op
e
rati
on
is
ne
eded
i
n
or
der
t
o
w
rite
al
l
copi
es
of
data.
T
hi
s
op
e
rati
on
re
qu
i
re
acce
ss
in
al
l
n
cop
ie
s
of
data,
if ther
e
are
n c
op
ie
s
(
a c
opy i
n
eac
h replic
a)
in the sy
stem
[6,
11
,
12]
.
RO
WA
will
translat
es
a
lo
gi
cal
read
operat
ion
on
a
data
i
tem
into
one
physi
cal
read
on
a
ny
of
it
s
rep
li
ca
an
d
tra
ns
la
te
s
al
l
it
s
l
og
ic
al
w
rite
s
operati
on
to
ph
ys
ic
al
wr
it
es
operati
on
on
e
at
each
re
plica
[
1
,
12
]
.
The
acce
ss
to
each
re
plica
s
will
be
synch
r
on
iz
e
by
the
m
ai
n
con
c
urre
nc
y
con
tr
oller
th
us
m
akes
this
protoc
ol
is
equ
ivale
nt
with
serial
execut
ion
w
her
e
eac
h
rep
li
ca
that
update
the
data
it
e
m
wil
l
up
dat
e
al
l
of
it
s
cop
ie
s
or
none
at
al
l.
RO
WA
is
go
od
for
the
e
nv
ir
on
m
ents
w
her
e
the
data
is
m
os
tl
y
read
-
only
be
cause
it
pro
vide
s
rea
d
op
e
rati
on
with
a
hig
h
de
gree
of
avail
abili
ty
at
low
com
mu
nicat
io
n
over
head
[11
,
12]
.
Howev
e
r,
the
wr
it
e
op
e
rati
on
has
ver
y
high
over
head
as
al
l
repl
ic
as
m
us
t
be
updated
sim
ult
aneously
and
c
annot
be
exec
uted
at
the
fail
ure o
f
a
ny
co
py.
T
his
r
esulte
d
i
n
a
n
i
m
balance
of
da
ta
avail
abili
ty
and
c
omm
un
icati
on
c
os
t of
re
ad
a
nd
wr
it
e
operati
ons.
T
he
rea
d
op
e
rati
ons
ha
ve
hig
h
a
vaila
bili
ty
and
lo
w
com
m
un
ic
at
i
on
c
os
t
w
hile
wr
it
e
op
e
rati
ons
i
n
t
he
oth
e
r
hand
hav
e
lo
w
a
vaila
bili
ty
with
hi
gh
e
r
c
omm
un
ic
at
ion
c
os
t
[11
,
12]
.
Des
pite
offer
i
ng
a
strong
co
ns
i
ste
ncy
an
d
a
high
de
gree
of
fau
lt
tolera
nt,
this
te
chn
i
que
s
increases
res
pons
e
ti
m
e
and
lo
w
acce
ss
c
oncu
rrency
[13
,
14]
.
The
ot
her
si
gn
i
ficant
draw
backs
of
RO
WA
a
re
this
te
chn
i
qu
e
is
ri
gid
i
n
sel
ect
ing
it
s r
e
ad
a
vaila
bili
ty
an
d i
na
dequate
in
the
ab
il
it
y o
f
tolerat
i
ng com
m
un
ic
at
ion
fa
il
ur
es
[6
]
.
2.2.
Qu
orum
cons
en
sus
(Q
C)
or
votin
g
The
Quorum
Con
se
nsus
(
Q
C)
m
e
tho
d
ge
ner
al
ly
al
low
wr
it
es
ope
rati
on
t
o
be
rec
orde
d
on
ly
at
a
subset
(a
w
rite
quorum
)
of
the
up
sit
es,
on
co
nd
it
io
n
that
reads
ope
rati
on
is
m
ade
to
query
a
s
ubset
(
a
read
quorum
)
sp
eci
f
ic
al
ly
pr
oved
will
ov
e
rlap
wi
th
w
rite
quoru
m
[1
1].
The
re
ad
op
e
rati
on
w
il
l
be
able
to r
e
tur
n
it
s
m
os
t
recently
wr
it
te
n
value
wh
e
ne
ver
the
quor
um
intersect
ion
c
onditi
on
i
s
m
et
and
will
be
sai
d
as
ha
ve
vote
d
for
it
,
giv
in
g
the
QC
m
et
ho
d
al
te
rn
at
ive
na
m
e,
vo
ti
ng.
Th
ere
are
tw
o
sta
te
s
of
voti
ng,
that
is
it
is
ei
th
er
sta
t
ic
or
dynam
ic
[11,
12
]
.
In
sta
ti
c
QC
sta
te
,
the
y
are
identifie
d
by
vo
te
s
t
hat
are
assig
ne
d
duri
ng
t
he
syst
em
sta
rt
up
ti
m
e
wh
il
e
for
dynam
ic
st
at
e
is
wh
en
the
sit
e
are
able
t
o
recon
fig
ur
e
it
s
qu
or
um
sp
eci
ficat
ion
in
re
sp
ons
e
to an
y
syst
e
m
s ev
ents
su
c
h
a
s
f
ai
lure, l
oa
d
c
ha
ng
e
s
or
else
[11
,
12
]
.
Like
R
O
WA
,
QC
m
et
ho
d
al
s
o
unde
rgo
fe
w
enh
a
ncem
ent
to
ov
e
rc
om
e
it
s
wea
kness.
Th
eref
or
e
eac
h
QC
m
et
ho
d
util
iz
es
diff
ere
nt
quorum
m
e
m
ber
s
hip
desi
gns,
va
ryi
ng
f
r
om
the
si
m
plest
nu
m
ber
of
quor
um
m
ajo
rity
to
an
unam
big
uous
m
e
m
ber
sh
ip
li
sti
ng
of
ea
ch
possible
quor
um
.
QC
was
first
a
pp
l
ie
d
in
the
rep
li
cat
ed
databases
by
usi
ng
Un
if
orm
Ma
j
ori
ty
Qu
or
um
Con
sens
us
Me
thod
[11
,
12]
.
This
m
et
ho
d
giv
e
s
adv
a
ntage
s
on
the
flexibili
ty
t
o
both
sit
e
and
able
to
tolerat
e
network
fail
ur
es
ho
we
ver
,
it
req
uire
d
high
read
and up
date co
s
ts [11
,
12]
.
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
Novel
ty
circula
r n
ei
ghborin
g
t
echn
i
qu
e
usi
ng
rea
ct
iv
e fa
ult t
olera
nce
meth
od
(
Ah
mad S
hukri
M
ohd
N
oo
r
)
5213
2.3.
Tw
o
-
repl
i
ca
dis
tribu
tio
n t
ec
hnique
(
TRDT)
Sh
e
n.
Che
n.
Z
heng
an
d
S
hi
[
12
]
ha
ve
pro
pose
d
Tw
o
-
Re
pl
ic
a
Distribu
ti
on
Tech
nique
(
TRDT)
that
introd
uced
eac
h
no
de
m
us
t
has
eq
ual
capaci
ty
of
sto
rag
e
a
nd
al
l
data
ha
ve
two
rep
li
cas
on
dif
fer
e
nt
no
des
a
s
well
as
al
l
no
des
ha
ve
tw
o
data
rep
li
cas
[
11
,
12]
.
For
N
nu
m
ber
of
nodes,
it
is
div
id
ed
to
n
set
of
nodes
(
N
=
2
n
)
wh
e
r
e
each
of
the
set
com
pr
ise
s
of
tw
o
nodes
.
The
m
ai
n
dr
a
wb
ac
k
f
or
this
te
chn
iq
ue
is
TRDT
increases
resou
rces
us
e
d
w
he
n
ad
ding
sec
ond
re
plica
to
the
serv
e
rs.
TR
D
T
al
so
lim
i
ts
the
op
e
rati
on
if
on
e
of
the
re
plica
set
is
absen
t,
le
a
vi
ng
t
he
no
de
e
xpose
d
t
o
do
ub
l
e
fail
ur
e
i
n
cas
e
of
both
r
epli
cas
are
da
m
aged
or
lost [
11,
12
]
.
2.4.
Neig
hb
or r
epli
ca
f
ailure
reco
very
(
N
R
FR)
A.
S
M
ohd
Noor
pro
po
se
d
N
ei
ghbor
Re
plic
a
Fail
ur
e
Re
co
ver
y
(
NRFR)
[
15
]
that
ba
sed
on
NRD
T
log
ic
al
struct
ure
for
data
re
pl
ic
at
ion
.
NRFR
log
ic
al
ly
or
ga
nized
in
t
he
f
orm
of
two
dim
ensio
nal
n
x
n
gr
i
d
structu
re
w
hen
there
are
node
s
in
th
e
syst
e
m,
N
=
n2
[
16
]
.
Each
of
the
nodes
owne
d
it
s
own
m
ast
er
da
ta
file
wh
e
re
up
dates
will
be
app
li
ed
on
the
pri
m
ary
data
file
befor
e
nav
i
ga
te
s
to
it
s
nei
ghbor
c
op
ie
s
within
a sepa
rate t
ra
nsa
ct
ion
s
[
15,
16]
.
2.5. Neig
hb
or
repl
ic
at
io
n
on grid
(
NRG)
Neig
hb
or
Re
plica
ti
on
of
Gr
i
d
has
been
int
rod
uced
by
N.
Ah
m
ad
[11]
by
app
ly
in
g
ne
ighborin
g
te
chn
iq
ues
in
a
two
-
dim
ensi
on
al
gri
d
str
uc
ture.
All
the
sit
es
in
this
m
e
chan
ism
is
logi
cal
ly
or
ga
nize
d
in
the form
o
f
n ×
n
st
ru
ct
ur
e
, by ha
ving a m
ast
er
data it
e
m
stor
ed
in eac
h
sit
e
s.
Let
r
de
no
te
a
s
read
quorum
,
w
de
note
d
a
s
wr
it
e
qu
oru
m
and
LB
is
the
total
num
ber
of
vote
s,
the consist
en
cy
of
data ob
j
ect
will
b
e c
on
t
ro
ll
ed by en
surin
g t
her
e is a
no
ne
m
pt
y i
ntersecti
on w
it
h:
a.
1
≤
≤
B
,
1
≤
≤
B
b.
+
=
B
+
1
The
com
plexity
of
m
ai
ntaini
ng
c
op
ie
s
at
di
ff
ere
nt
sit
es
ga
ve
a
chall
en
ge
d
f
or
NRG
t
o
ha
ndle
it
s
locking
inf
or
m
at
ion
es
pecial
ly
in
the
current
struct
ure o
f dist
rib
uted
co
m
pu
ti
ng.
3.
CIRC
ULA
R NEIGHB
ORING
MO
DEL
AND ME
TH
OD
This
sect
io
n
will
discuss
e
d
the
new
ly
pr
o
pose
d
sync
hrono
us
Ci
rcu
la
r
Neig
hbori
ng
rep
li
cat
io
n
te
chn
iq
ue,
C
N
R t
hat u
se r
eac
ti
ve
m
et
ho
d for f
a
ult t
oleran
c
e p
olicy
[
17]
. T
he
ne
ig
hbor
i
ng tech
nique adop
te
d
a
neig
hbori
ng
re
la
ti
on
sh
i
p
that
is
identic
al
to
a
ci
rcle
where
a
nu
m
ber
of
re
plica
s
are
relat
ed
to
eac
h
oth
e
r
bac
k
to
back.
He
nce
,
each
re
plica
has
two
neig
hbors,
the
one
pre
cedin
g
it
and
the
one
fo
ll
ow
i
ng.
This
relat
io
ns
hi
p
is t
ru
e
f
or
re
plica
s of three
or
m
or
e, whil
st t
wo r
e
plica
s w
il
l solel
y be
neig
hbors to
one a
nothe
r
[
18]
.
A
set
of
re
plica
s
is
def
ine
d
wh
e
re
N
is
the
total
nu
m
ber
of
rep
li
cas.
As
il
lustrate
d
in
a
ci
rcle,
for
the
ba
se
case
of
three
re
plica
s,
eac
h
base
case
of
t
hr
ee
r
eplic
as,
eac
h
m
e
m
ber
of
the
re
plica
will
ha
ve
t
w
o
neig
hbors
i.e.
has
an
d
as
neighb
or
s
.
For
e
xam
ple,
if
on
e
m
or
e
rep
li
ca
w
il
l
be
add
e
d
to
the
set
,
the
re
plica
is
inserted
to
fol
low
the
prev
iou
s
re
plica
.
Th
us
,
it
brea
ks
up
t
he
ne
ighbor
relat
ion
sh
i
p
betwee
n
a
nd
conseq
ue
ntly
,
r
eplace
d
it
with
new
ly
a
dd
e
d
r
eplic
a.
By
it
erati
on
,
a
dd
i
ng
m
or
e
rep
li
cas
to
the
set
will
un
derg
o
the
s
am
e
pr
oce
ss
an
d
c
onseq
ue
nce.
T
he
refore
,
it
is
pro
ven
that
for
a
ny
nu
m
ber
of
rep
li
c
as
of
th
ree
or
m
or
e,
each
of
them
has
tw
o
nei
ghbo
rs
prece
di
ng
a
nd
f
ollowi
ng
them
,
in
a
ci
rcu
la
r
neighb
or
i
ng
relat
ion
s
hip.
The
CNR
lo
gical
structur
e
de
sign
e
xam
ple
can
be
sho
wn
as
sh
own
in
F
igure
1
.
Assu
m
ing
that
the
pr
im
ary
data file
f
1
is l
oc
at
ed
on t
he
se
rv
e
r
sit
e
n
1
, the
prim
ary data file
f
2
locat
ed
on the
serve
r
sit
e
n
2
an
d so
on.
Figure
1
.
Th
e
a
rr
a
ng
em
ent of
pr
im
ary rep
li
cas in C
NR w
it
h 6 nodes
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.
6
,
Dece
m
ber
201
9
:
5211
-
5217
5214
Each
sit
e
hold
s
three
c
op
ie
s
of
un
i
qu
e
dat
a
with
one
of
them
is
it
s
ow
n
pri
m
ary
data
an
d
ea
c
h
ind
ivi
du
al
data
will
has
t
hr
ee
cop
ie
s
any
wh
e
re
on
t
he
syst
e
m
.
W
ei
gh
a
gai
ns
t
f
ull
data
re
plica
ti
on
,
t
his
new
ly
pro
po
se
d
te
c
hniqu
e
offe
rs
m
ini
m
iz
ation
on th
e stor
a
g
e ca
pac
it
y wit
ho
ut
a
ba
ndonin
g
syst
e
m
av
ai
la
bility
.
Quorum
is
a
num
ber
of
re
plica
s
that
li
ste
d
i
n
a
n
a
gr
eem
ent
for
a
ny
opera
ti
on
to
ta
ke
pla
ce
on
them
.
The
m
echan
ism
m
us
t
be
able
to
m
anag
e
two
op
e
rati
ons
of
rea
d
a
nd
wri
te
in
the
case
of
bo
t
h
operati
on
s
run
con
c
urre
ntly
in
order
t
o
av
oid
any
co
nf
li
ct
s
excep
t
for
rea
d
-
rea
d
ope
rati
ons.
The
refo
re,
da
ta
synchro
nizat
io
n
is
a
cr
ucial
pa
r
t
for
a
re
plica
tio
n
te
ch
nique.
Data
can
be
sy
nchr
on
iz
e
e
ff
ic
ie
ntly
by
ad
op
t
ing
t
hese
t
wo
r
ules
for quo
ru
m
v
ot
ing
:
a.
Re
ad
per
im
et
e
r,
R
and
w
rite
per
im
et
er,
W
m
us
t
be
gr
eat
er
than
certai
n
non
-
neg
at
i
ve
nu
m
ber
of
vote
s
,
R +
W
>
v
b.
Wr
it
e
per
im
et
e
r
m
us
t be
gr
eat
er th
a
n n
on
-
ne
gative
nu
m
ber
of votes
div
ide
b
y
2,
W
>
v
/
2
The
sec
ond
r
ul
es
on
ly
ap
plica
ble
if
the
ve
rsion
n
um
ber
is
r
equ
i
red
to v
er
i
fy
the
m
os
t
cur
ren
t
co
py
of
the
re
plica
.
T
hi
s
two
r
ules
ar
e
i
m
po
rta
nt
to
avo
i
d
c
on
c
urre
nt
data
w
rite
s
and
rea
d
as
we
ll
as
wr
it
e
op
e
r
at
ions
occurre
d
on
t
he
sam
e
data
at
the
sam
e
tim
e
.
As
the
res
ult,
rea
d
sc
or
es
al
ways
re
flect
th
e
m
os
t
recent
wr
it
e
because
rea
d
quor
um
con
sist
of
at
le
ast
on
e
rep
li
ca
that
w
as
involve
d
i
n
the
m
os
t
recent
w
rite
operat
ion
s
.
Fo
r
insta
nce,
quor
um
is
m
od
el
le
d
as
a
set
of
rep
li
cas;
C
=
{
C
1
,
C
2
,
C
3
…
C
n
}
w
her
e
n
=
1,
2,
3,
….
,i
are
c
al
le
d
the
se
qu
e
nce
nu
m
ber
s
of
th
ese
re
plica
s.
Each
rep
li
ca,
C
i
m
anag
e
a
set
of
data.
I
n
C
NR,
as
di
scusse
d
pr
e
viously
ha
ve
on
ly
tw
o
(
2)
neig
hbors
to
e
ach
re
plica
an
d
only
three
(3)
re
plica
s
co
ntain
the
sam
e
c
op
y
of
a
data
it
e
m
.
To
en
sure
on
e
c
op
y
se
rial
iz
abili
ty
in
CNR
po
li
cy
,
read
ope
rati
on,
r
quoru
m
will
be
def
i
ned
a
s
r
=
1
a
nd
f
or
wr
it
e
ope
rati
on,
w
de
fine
d
a
s
w
=
3.
F
r
om
Figure
2
,
a
ny
two
a
djacent
rep
li
cas
ca
n
be
a
read
quorum
.
The
wr
it
e
quorum
is
al
l
three
re
plica
s
that
s
ha
res
the
m
utu
al
data
with
the
two
re
plica
s
in
re
ad
quorum
.
Con
si
der
a
pri
m
ary
da
ta
it
e
m
of
d
2
.
Ba
sed
on
CN
R
ru
le
s,
ass
umi
ng
t
hat
n
=
3
wh
e
re
eac
h
repl
ic
a
get
on
e
vo
te
a
nd
al
l
three
rep
li
ca
con
ta
in
the
sam
e
cop
ie
s
of
pr
im
ary
data
d
2
,
d
1
a
nd
d
3
ho
ld
s
the
co
pi
es
of
the
pr
im
ary
data.
Thus,
in
order
to
pe
rfor
m
read
on
d
2
pri
m
ary
data,
the
valid
qu
orum
set
are
{
d
1
},
{
d2
}
,
{
d3
}
,
{d
2
,
d
1
}
, {d
1
,
d
3
}
, {d
2
,
d
3
}
an
d
{d
2
, d
1
, d
3
}
. T
he write
quor
um
is {
d
1
,
d
2
,
d
3
}.
Figure
2
.
Repli
ca o
rg
a
nizat
ion o
f
CNR
of siz
e
n
(a
ny intege
r)
In
each
CNR
re
plica
,
c
oor
din
at
in
g
al
gor
it
h
m
con
tr
ols
the
pri
m
ary
re
plica
w
hile
co
operati
ng
al
gorithm
deals
with
no
n
-
pr
i
m
ary
rep
li
ca.
Wh
e
n
t
he
cl
ie
nt
se
nd
s
it
s
re
qu
e
st,
CNR
tra
ns
act
io
n
m
anag
er
will
m
anag
e
the
re
qu
e
st
from
c
lients
an
d
locat
es
the
pr
im
ary
rep
li
ca.
Pr
im
a
ry
rep
li
ca
will
execu
te
co
ord
inati
ng
al
gorithm
up
on
recei
ving
procedu
re
cal
l
from
transacti
on
m
anag
er
.
T
he
se
process
es
a
re
di
vid
e
d
int
o
fe
w
ph
a
ses.
I
n
the
first
phase,
the
pr
im
ary
rep
li
ca
asks
relat
ed
r
eplic
as
to
cast
vo
te
in
order
t
o
f
or
m
a
qu
o
rum
fo
r
the
operati
on.
If
quorum
is
form
ed,
it
will
return
the
value
of
1
t
o
tra
ns
act
ion
m
anag
e
r
or
el
se
it
will
return
0
and
ab
or
t
t
he
op
e
rati
on.
In
s
om
e
cases,
the
quorum
will
return
-
1
in
the
case
of
par
ti
a
l
com
m
itm
ent
fro
m
the
relat
ed
re
plica
s.
D
ur
i
ng
the
seco
nd
ph
a
se,
the
t
ra
ns
act
io
n
m
anag
er
will
ask
quorum
to
co
m
m
i
t
the
operati
ons
i.e.
lock
the
data.
I
f
there
is
any
par
ti
al
com
m
itm
ent
i
nvolv
e
d,
pr
im
ary
rep
li
ca
ca
ptures
the
detai
ls
of
wh
ic
h
re
plica
issued
pa
rtia
l
com
m
itm
ent
and
will
us
e
th
e
m
to
so
lv
e
in
c
onflic
t
resol
ution.
Durin
g
operat
ion
exec
utio
n,
cooper
at
in
g
al
gorithm
a
t
the
non
-
pri
m
a
ry
rep
li
cas’
s
ide
will
per
f
or
m
the ope
rati
on a
nd unloc
k
t
he data
whe
n
it
is
done.
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
Novel
ty
circula
r n
ei
ghborin
g
t
echn
i
qu
e
usi
ng
rea
ct
iv
e fa
ult t
olera
nce
meth
od
(
Ah
mad S
hukri
M
ohd
N
oo
r
)
5215
4.
CIRC
ULA
R NEIGHB
ORING
A
NA
L
YSIS A
N
D
E
VAL
UA
TIO
N
This
sect
io
n
w
il
l
fu
rthe
r
a
naly
ze
t
he
avail
a
bili
ty
of
the
ne
wly
pro
po
se
d
rep
li
cat
ion
te
chn
i
qu
e
[
19
]
.
Av
ai
la
bili
ty
of
a
syst
em
ref
ers
to
the
pro
bab
il
it
y
of
a
syst
e
m
is
com
ple
te
ly
wo
r
ki
ng
over
a
pe
rio
d
of
op
e
rati
ng
tim
e
[2
0].
Ta
ble
2
sh
ows
f
or
eac
h
4
te
chn
iq
ues
the
num
ber
of
r
eplic
as
require
d
for
a
data
it
e
m
or
obj
ect
in
f
our
syst
e
m
scenario
s
with
diff
e
r
ent
nu
m
ber
of
n
nodes
ass
um
ing
that
the
rep
li
cas
are
unde
r
a
norm
al
ci
rcu
m
sta
nces
wit
hout
any
fail
-
sto
p
fail
ur
e
.
As
sum
e
that
du
ri
ng
fail
-
sto
p
fail
ure,
on
e
re
plica
is
not
avail
able,
as
per
li
st
ed
on
Table
3.
Final
ly
,
Table
4
re
cords
al
l
a
vaila
bili
ty
tim
es
fo
r
e
ach
te
c
hniqu
e
s
dep
e
ndin
g on the tec
hniq
ue
ar
chite
ct
ur
e
base
d on Ta
ble 1 c
om
po
ne
nts’
a
va
il
abili
ti
es.
Table
1
.
T
he
n
i
ne
c
om
po
ne
nts
of inter
de
pend
ent ser
ve
rs
a
nd it
s av
ai
la
bili
ti
es
Co
m
p
o
n
e
nt
Av
ailab
ility
W
eb
0
.95
Ap
p
licatio
n
0
.95
5
Databas
e
0
.95
DNS
0
.97
Firewall
0
.96
Switch
0
.97
Data Cent
re
0
.95
Ap
p
licatio
n
2
0
.95
Manag
er
0
.99
Total Av
ailab
ilit
y
0
.69
5
6
Table
2
.
T
he
c
om
par
ison o
f
t
he
siz
e
of r
e
plica
s un
der
dif
fe
ren
t set
of
n nodes
for
m
ulti
ple r
eplic
at
ion
tec
hniqu
e
Rep
licatio
n
Techn
iq
u
e
Nu
m
b
e
r
o
f
r
ep
lica
f
o
r
each d
ata/o
b
ject
n
=
1
2
n
=
2
4
n
=
3
5
n
=
4
0
ROWA
12
24
35
40
TRDT
2
2
2
2
NRFR
5
5
5
5
CNR
2
2
2
2
Table
3
.
T
he
c
om
par
ison o
f
t
he
siz
e
of r
e
plica
s dur
i
ng f
ai
l
-
stop fail
ure f
or
m
ult
iple replic
at
ion
tec
hn
i
qu
e
Rep
licatio
n
T
echn
iq
u
e
Nu
m
b
e
r
o
f
r
ep
lica
f
o
r
each d
ata/o
b
ject
n
=
1
2
n
=
2
4
n
=
3
5
n
=
4
0
ROWA
11
23
34
39
TRDT
1
1
1
1
NRFR
4
4
4
4
CNR
2
2
2
2
Table
4
.
T
he
c
om
par
ison o
f
i
m
pr
ov
em
ents u
sin
g dif
fer
e
nt
rep
li
cat
ion m
od
el
for 9
no
des
Rep
licatio
n
T
echn
iq
u
e
Sy
ste
m
Availab
ilit
y
I
m
p
rov
e
m
en
t
(
%)
No
r
ep
licatio
n
0
.69
5
6
0
.00
0
ROWA
0
.91
0
3
3
0
.86
2
TRDT
0
.98
4
6
4
1
.54
7
NRFR
0
.99
7
8
4
3
.44
4
CNR
0
.99
9
5
4
3
.69
1
Re
su
lt
s
show
that
the
im
pr
ovem
ents
in
syst
e
m
avail
ability
in
9
no
des
var
ie
d
with
r
eplic
at
ion
te
chn
iq
ue.
T
he
diff
e
ren
t
am
ou
nt
of
re
plica
te
d
node
s
in
each
te
ch
nique
con
tri
bu
te
s
a
s
the
m
ai
n
factor
in
pro
vid
in
g
a
hi
gh
ly
a
vaila
ble
syst
e
m
.
Fr
om
Table
2
a
nd
T
able
4,
it
is
cl
ear
that
rep
li
c
at
ion
te
ch
ni
que
with
CNR
achieve
s
the
highest
sys
tem
avail
abili
t
y
with
the
lo
w
est
rep
li
cat
ion
req
uire
d
to
m
a
intai
n
it
s
avail
abili
ty
.
CNR
re
quires at
m
os
t
2
re
plica
s
in ord
e
r
t
o
search
f
or
the
n
odes
that ho
l
d
the
sam
e
data file
of
the
fail
e
d
node
wh
il
e
for
RO
WA
a
nd
NRF
R
req
uir
es
ad
diti
on
al
tim
e
t
o
recove
r
due
to
the
increm
ent
of
the
nu
m
ber
of
rep
li
cas
f
or
a
data
it
e
m
du
e
to
the
increase
of
num
ber
of
nodes
in
the
env
i
ronm
ent.
Wh
il
e
f
or
TR
DT
that
sh
are
s
the
sam
e
num
ber
of
r
eplic
as
with
C
NR
sc
or
es
lo
w
er
syst
em
avail
abili
ty
du
e
to
the
a
rr
a
ngem
ent
of
nodes
in
t
he
te
chn
i
qu
e
a
rc
hitec
ture.
F
urt
he
r
m
or
e,
in
the
e
ve
nt
of
fail
-
sto
p
fail
ur
e,
TR
DT
on
ly
has
one
r
eplic
a
le
ft
fo
r
rec
ove
ry,
thu
s
res
ulted
in
lowe
r
scor
e
of
rec
ov
e
r
y
la
te
ncy
wh
ic
h
aff
ect
e
d
it
s
syst
e
m
avail
a
bili
ty
.
Ther
e
f
or
e,
f
or
a
rec
ov
e
ry
pro
cedure,
it
can
be
deduce
d
tha
t
syst
e
m
avail
a
bili
ty
i
m
pr
ov
es
furthe
r
c
onsist
ently
to the n
um
ber
of r
e
plica
s in
t
he
syst
em
.
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.
6
,
Dece
m
ber
201
9
:
5211
-
5217
5216
Fr
om
the
resu
l
ts,
RO
WA
has
sho
wn
that
as
the
c
om
plexity
of
the
syst
em
increases,
t
he
te
c
hn
i
qu
e
beco
m
e
irrelev
ant
to
be
us
e
t
o
rec
over
t
he
syst
e
m
as
the
nu
m
ber
of
dat
a
file
gro
ws
as
with
the
num
ber
of
nodes
in
t
he
s
yst
e
m
.
Thu
s
,
it
lowe
rs
t
he
sy
stem
avail
abili
ty
.
Howe
ver,
t
echn
i
qu
e
s
that
ad
op
te
d
neig
hbori
ng
po
li
cy
that
is
NRFR,
TR
DT
and
C
NR
are
not
af
fected
with
the
i
ncr
ease
nu
m
ber
of
node
s
in
the
syst
e
m
.
CNR
le
ads
NRFR
a
nd
TRDT
i
n
te
rm
s
of
highe
r
degre
e
fau
lt
re
cov
e
ry
as
it
re
qu
i
res
lo
wer
re
plica
s
than
NR
FR
an
d
in the e
ve
nt
of
fail
-
sto
p
it
is
not l
i
m
it
ed
to on
ly
o
ne re
plica
av
ai
la
ble
per
da
ta
it
e
m
as in
T
RDT.
5.
CONCL
US
I
O
N
A
un
i
qu
e
ci
rc
ular
neig
hbori
ng
te
ch
nique
base
d
on
reac
ti
ve
fa
ult
tolerance
m
et
ho
d
has
bee
n
pr
ese
nted
.
T
he
te
ch
nique
prov
i
des
l
ow
e
r
data
re
plica
ti
on
c
om
par
ed
to
oth
e
r
sta
te
of
the
art
te
c
hn
i
qu
e
s
.
The
m
od
el
an
d
m
et
ho
do
l
ogy
of
the
te
ch
ni
qu
e
hav
e
bee
n
desc
ribe
d
al
ong
with
anal
ysi
s
and
eval
ua
ti
on
of
the tec
hniq
ue.
CNR
is
an
appro
ac
h
f
or
data
rep
li
cat
ion,
by
placi
ng
re
dundant
co
pies
of
data
in
a
ci
rcu
la
r
fo
rm
to
i
m
pr
ove
the
qual
it
y
of
ser
vi
ce
in
a
feasib
ly
cl
utter
distribu
te
d
e
nvir
onm
ent.
CNR
te
chn
i
qu
e
deter
m
ines
the
m
os
t
app
r
opriat
e
nei
ghbo
r
to
perform
data
rep
li
cat
ion
in
a
sm
al
l
qu
orum
siz
e
to
ex
ecute
the
proc
edures.
The
pe
rfo
rm
a
nce
analy
sis
of
CNR
com
par
ed
with
ot
her
te
chn
iq
ues
s
howe
d
that
CN
R
prov
i
des
a
bette
r
appr
oach
t
o
highly
avail
able
rea
d
an
d
wr
it
e
ope
rati
on
for
distri
bu
te
d
syst
em
.
CNR
offe
rs
lowe
r
com
m
un
ic
at
ion
co
st
to
re
plica
te
the
data
it
e
m
s
as
well
as
m
ai
ntaining
t
he
hi
gh
e
r
sc
ore
of
recove
ry
la
te
ncy
thu
s
r
es
ulted i
n hig
he
r
syst
em
avail
abili
ty
co
m
par
ed
with
o
t
her te
ch
niques.
Wh
il
e
c
om
pr
ehensi
ve
a
naly
sis
and
com
pari
so
ns
ha
ve
be
en
done
for
t
his
w
ork
set
ti
ng,
f
urt
he
r
exp
e
rim
ents
are
require
d
be
fore
a
com
plete
l
y
decisi
ve
eval
uation
of
t
he
pro
posed
al
gorithm
can
be
achieve
d.
Howe
ver,
the
pro
po
se
d
te
c
hniq
ue
s
howe
d
that
CNR
a
ble
to
i
ncr
ease
a
syst
e
m
avail
abili
t
y
by
m
ai
nta
inin
g
si
m
plest
rep
li
ca
distrib
utio
n
wh
ic
h
able
t
o
reduce
the
ti
m
e
ta
ken
t
o
rec
ov
e
r
a
fail
e
d
s
yst
e
m
co
m
par
ed
with
the
existi
ng
te
chn
i
qu
e
s.
The
se
reassurin
g
resu
lt
s
sho
w
that
the
te
chn
i
qu
e
pro
po
se
d
in
the
pa
per
a
ble
to
i
m
pr
ove fa
ult
m
anag
em
ent an
d co
ntri
bu
te
s
towa
rd
s
conti
nuous
ser
vice
de
li
ver
y i
n dist
ri
bu
te
d
syst
em
s.
ACKN
OWLE
DGE
MENTS
This
re
searc
h
is
fun
ded
by
Fund
am
ental
Re
search
Gr
a
nt
Sc
hem
e
(F
R
GS
)
wi
th
the
Re
f:
FRGS/1/
2018/
ICT0
4/UMT/
02/2.
FR
GS
is a
researc
h gr
a
nt
from
the Minist
ry o
f
E
du
cat
io
n (MOE
)
Ma
la
ysi
a
.
REFERE
NCE
S
[1]
Noor
A.S.
M.,
Zi
an
N.F.M.,
Shaiful
Bahri
F.N.
M.
,
"
Surve
y
on
rep
li
c
at
ion
te
c
hnique
s
for
distri
bute
d
s
y
st
em,"
Inte
rnational
Jo
urnal
of El
e
ct
ri
c
al
and
Comput
er
Engi
n
ee
ring
(
IJE
CE)
,
vol
.
9(2)
,
pp.
1298
-
1303
,
2019.
[2]
J.A.
Torke
stani,
"A
Highl
y
Re
li
able
And
Para
ll
eliz
abl
e
Dat
a
Distribut
ion
Sc
heme
For
Data
Grids
,
"
Fut
ure
Gene
ration
Com
pute
r Sy
st
ems
,
vol.
29(2)
,
pp
.
50
9
-
519,
2013
[3]
N.
Chera
gh
lou
,
et
a
l
.,
“
A
Survey
Of
Faul
t
To
leranc
e
Archi
t
ec
t
ure
In
Cloud
C
om
puti
ng
,
”
E
lse
vi
er
Journal
o
f
Net
work
and
Co
mputer
Appli
cati
ons
,
vol.
61
,
pp
.
81
-
92,
2016
[4]
Noor
A.S.M.,
D
eri
s
M.M.
,
“
Fail
ure
rec
over
y
m
ec
han
ism
in
nei
ghbor
rep
li
c
a
distri
buti
on
arc
h
itect
ur
e
,
”
Lect
ur
e
Note
s in
Comput
er
Scienc
e
.
Vol
.
6377
LNCS,
Iss
ue
M4D
,
2010
,
Pages
41
-
48.
Springer
Na
ture.
20
10.
[5]
Y.
M
Te
o
a
t
el.
“
Cost
-
Perform
a
nce
of
Faul
t
Tolera
nc
e
in
Cloud
Com
puti
ng”,
Int
ernati
onal
Conf
e
renc
e
on
Ad
vance
Computing
and
Appl
ic
a
ti
ons (
Spec
ia
l
Iss
ue
o
f Jo
urnal
of
S
ci
en
ce
and
Technol
og
y)
,
vo
l.
49,
pp
61
-
73,
2011
.
[6]
A.
A
.
Hela
l
,
A.
A.
Hedda
y
a
an
d
B.
B.
Bharga
v
a,
R
epl
i
cat
ion
Tec
hnique
s
In
Distribute
d
Syst
ems
,
2nd,
Kluwer
Aca
demic
,
1996
.
[7]
Mohd.
Noor,
A.S.,
Ma
t
Deri
s,
M.
“
Ext
ende
d
hea
rtb
eat
m
echani
sm
for
fau
l
t
detec
t
ion
serv
ic
e
m
et
hodo
lo
g
y
”
Comm
unic
ati
ons i
n
Comput
er
an
d
Information
S
c
ie
nc
e
Vol
.
63
,
p
p.
88
-
95
.
Spring
er
Natur
e. 2009
[8]
R.
V.
Rene
ss
e
an
d
R.
Guerra
oui
.
,
“
Repl
ic
at
ion
Techni
ques
for
Availabil
i
t
y
,
”
In
B.
Charr
on
-
Bost,
R.
Pe
don
e,
an
d
A.
S
chi
p
er
(
Eds.
)
:
Re
plicati
on
,
L
ec
ture
Note
s
in Com
pute
r Sc
i
en
ce
,
vo
l.
5959,
p
p
19
-
40,
2010
[9]
A.
Nata
nzon
,
E
.
Bac
hm
at
,
“
D
y
na
m
ic
S
y
nchr
on
ous/As
y
n
chr
ono
us
Repl
ic
a
ti
on”
,
ACM
Tr
ans
-
ac
ti
on
on
Storage
,
v
ol.
9(
3
)
,
2013
.
[10]
Mohd
Noor
A.S.,
Yunus
F.,
Ma
m
at
R.
,
Sir
aj
udd
in
E.
A
.
,
Ma
t
Z
in
N.F.
,
"
Enh
anci
ng
ava
i
la
bi
li
t
y
o
f
m
ari
ne
big
data
rep
ositor
y
with
a
new
fau
lt
t
ole
ran
ce
t
ec
hni
que
,”
Journal
of
Tele
communi
cat
ion
,
El
e
ct
ron
ic
and
Computer
Engi
ne
ering
,
vo
l.
9(3
-
5
Spec
ia
l
I
ss
ue)
,
pp.
83
-
87,
2017.
[11]
A.
Noraz
ia
h
,
“
Mana
ging
Repli
ca
t
ion
and
Tr
an
sac
ti
ons
Us
ing
Neighbour
Replication
on
Dat
a
Grid
Data
b
ase
Design,
”
Ph
.
D. The
sis
.
Univ
ersi
ti
Ma
lay
sia
T
erengganu,
2007
.
[12]
Deri
s,
M.M.
"Ef
fic
i
ent
Acc
ess
of
Repl
icate
d
Da
ta
in
Distribut
ed
D
at
ab
ase
S
y
stems
.
"
Ph.D.
The
sis
,
Univer
siti
Putra
Malay
s
ia,
2001
.
[13]
Dabrows
ki
C.
,
Re
li
ab
il
i
ty
in
g
rid
computi
ng,
Concurrency
Comput
ati
on:
P
racti
c
e
and
Experie
nc
e
,
W
ile
y
Inte
rSci
enc
e
,
20
09.
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
Novel
ty
circula
r n
ei
ghborin
g
t
echn
i
qu
e
usi
ng
rea
ct
iv
e fa
ult t
olera
nce
meth
od
(
Ah
mad S
hukri
M
ohd
N
oo
r
)
5217
[14]
D.M.
Bui
,
T
.
H.
The
,
S.Y.
Lee,
F
uzz
y
faul
t
detec
tion in
IaaS
Cloud
Computing
,
ACM
IMCO
M
,
2016
.
[15]
Mohd
Noor,
A.S.,
Deri
s,
M
.
a
nd
Sam
an,
M.Y.,
"Co
-
Exi
st
ance
Neighbour
hood
m
odel
ava
i
la
bi
lit
y
p
red
i
ct
ion
fo
r
Cloud
Infra
struc
ture
as
a
serv
ice
(Ia
a
S)
opt
imiza
t
ion
in
interd
epe
nden
t
envi
ro
nm
ent
,
"
Inte
rna
ti
onal
Journal
of
Mac
hine Learni
ng
and
Computi
ng
,
v
ol
.
4
,
pp
.
85
,
2014
[16]
Mam
at
R.
,
Der
is
M.
and
Jal
il
M.
,
"N
ei
ghbor
Replica
Distri
buti
on
Te
chn
iq
ue
for
Cluste
r
Serve
r
S
y
s
te
m
s
,
"
Malay
sian
Journal
of
Comput
er
S
ci
en
ce,
v
ol
.
17
(
2
),
2004
.
[17]
Siva
S.S.
and
Babu
K.S.
,
"
Surve
y
of
f
aul
t
to
le
r
ant
techniqu
es
for
grid
,
"
Computer
Sci
en
ce
R
evi
ew
,
El
se
vier
Inc
.
,
vol.
4(2)
,
pp
.
10
1
-
120
,
2010
.
[18]
Mam
at
,
A,
Deri
s,
et
al
.
,
“
Mana
ging
Data
Us
ing
Neighbor
Repl
i
ca
t
ion
on
Tri
an
gula
r
-
Grid
Struc
ture
,
”
v
ol.
3994.
pp
.
1071
-
1077
,
2006
[19]
Rene
ss
e,
R.
V.
an
d
Guerra
oui,
R.
,
"Repl
ic
a
ti
on
T
e
chni
ques
for
Availabil
i
t
y
".
In
B
.
Charr
on
-
Bost,
R.
Pe
don
e,
and
A.
S
chi
p
er
(
Eds.
)
:
Re
pli
ca
ti
on
,
L
ec
ture
Note
s
in Com
pute
r Sc
i
en
ce
,
v
ol
.
5959
,
pp
.
19
-
40
,
2010
.
[20]
Schm
idt
K.
High
,
A
vai
lab
il
i
ty an
d
Disas
te
r R
e
co
ve
ry:
Con
ce
p
ts, Design,
Impl
eme
ntat
ion
,
Springer
-
Verl
ag
,
2006
.
Evaluation Warning : The document was created with Spire.PDF for Python.