Int
ern
at
i
onal
Journ
al of Ele
ctr
ic
al
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
8
, No
.
6
,
Decem
ber
201
8
, p
p.
5333
~
5341
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v8
i
6
.
pp
533
-
5431
5333
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Perform
ance B
en
chmark
ing of
Key
-
Valu
e Store
NoSQL
Databas
es
Omor
uy
i
Ose
mw
egie, Ke
nn
edy Ok
ok
p
uj
i
e, N
sik
an N
kordeh,
Charle
s
Nd
u
jiub
a,
Samuel
Joh
n,
Uz
airue S
tanle
y
Depa
rtment
o
f
E
le
c
tri
c
al a
nd
Inf
orm
at
ion
Eng
ineeri
ng,
Coven
ant
Univer
sit
y
,
Nige
ria
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Sep
6,
2017
Re
vised
Ma
y
3, 20
18
Accepte
d
J
un
11, 201
8
Inc
rea
sing
req
u
i
rement
s
for
sca
la
bil
i
t
y
and
elasti
ci
t
y
of
data
storage
for
web
appl
i
ca
t
ions
has
m
ade
Not
Stru
ct
ure
d
Quer
y
L
angua
ge
NoS
QL
da
ta
bas
es
m
ore
inva
lua
b
le
to
web
dev
el
op
ers.
One
of
such
NoS
QL
Data
ba
se
soluti
ons
is
Redi
s.
A
bud
ding
al
te
rn
ative
to
Redi
s
dat
aba
se
is
the
SS
D
B
dat
aba
se
,
which
is
al
so
a
ke
y
-
v
al
u
e
stor
e
but
is
disk
-
b
ase
d.
Th
e
ai
m
of
t
his
rese
ar
ch
work
is
to
b
enchm
ark
both
da
t
aba
ses
(Red
is
a
nd
SS
DB)
using
the
Y
ahoo
Cloud
Serving
Benc
hm
ark
(YCS
B).
YCS
B
is
a
pla
tform
tha
t
h
as
bee
n
use
d
to
compare
and
benc
hm
ark
sim
ilar
NoS
QL
dat
ab
ase
s
y
st
ems
.
Both
dat
ab
ase
s
were
giv
en
va
r
ia
bl
e
workloads
to
id
ent
if
y
th
e
throughput
o
f
al
l
give
n
oper
ations.
The
result
s
obta
in
ed
show
s
tha
t
SS
D
B
give
s
a
bet
t
er
throughput
for
m
aj
ority
of
o
per
ations t
o
Red
is’s pe
rform
anc
e
.
Ke
yw
or
d:
NoSQL
Re
dis
SSD
B
data
bas
e
Yaho
o
Cl
ou
d Ser
ving
Be
nch
m
ark
(YC
SB)
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
:
Kenne
dy Ok
okpuj
ie
,
Dep
a
rtm
ent o
f El
ect
rical
an
d I
nfor
m
at
ion
E
nginee
rin
g,
Cov
e
na
nt Univ
ersit
y,
KM 10
I
dir
oko
Roa
d,
Ota,
O
gun St
at
e,
Nige
r
ia
.
Em
a
il
: ok
okpu
j
ie
ke
nn
e
dy@c
ov
e
na
ntunive
rrsi
ty
.ed
u.n
g
1.
INTROD
U
CTION
Ther
e
is
an
i
ncr
easi
ng
prol
iferati
on
of
N
ot
Stru
ct
ur
e
d
Qu
e
ry
Lan
gu
a
ge
(
N
oSQL
)
databases
.
Am
on
gs
t
their
key
adv
a
ntag
es
is
the
pr
om
i
se
of
faster
a
nd
eff
ic
ie
nt
pe
rfor
m
ance
than
the
le
gacy
Re
la
ti
on
al
Database
Ma
na
gem
ent
Syst
e
m
s
(RDBMS)
[1
]
.
NoSQL
da
ta
bases
are
al
so
ta
il
or
fitt
ed
to
the
fast
gr
ow
i
ng
world
of
cl
ou
d
com
pu
ti
ng,
al
lowing
fo
r
m
assive
scal
in
g
“
on
dem
and
”
(e
la
sti
ci
t
y)
an
d
s
i
m
plifie
d
ap
plica
ti
on
dev
el
op
m
ent
[
2].
H
ow
e
ve
r,
t
her
e
a
re
w
ords
of
ca
ution
t
o
the
ba
ndwa
gon
of
ad
opti
on
in
big
data
an
d
we
b
app
li
cat
io
n
de
velo
pm
ent
no
t
ing
t
hat
not
a
ll
No
S
QL
dat
abases
a
re
c
re
at
ed
al
ike
where
pe
r
form
ance
is
con
ce
r
ned
[3
]
.
[4
]
asserts
that
since
N
oSQL
so
luti
ons
are
not
m
a
ture
an
d
are
pro
gr
e
ssin
g
at
dif
fer
e
nt
sp
eed
s,
database
a
dm
i
nistrato
rs
hav
e
to
ch
oose
ca
re
fu
ll
y
betwee
n
NoSQL
an
d
re
la
ti
on
al
databa
ses
acco
rd
i
ng
to
thei
r
sp
eci
fic
nee
ds
in
te
rm
s
of
co
ns
ist
e
ncy,
pe
rfor
m
ances,
sec
ur
it
y,
scal
abili
ty
,
costs
an
d
oth
er
non
-
f
un
ct
io
nal
crit
eria.
W
it
h
t
he
s
ubsta
ntial
num
ber
of
ope
n
-
s
ource
an
d
read
il
y
a
va
il
able
N
oSQL
syst
e
m
s,
we
b
app
li
cat
io
ns
dev
el
op
e
r
al
so
exp
e
rience
t
he
head
ac
he
of
sel
ect
ing
am
on
gs
t
su
c
h
NoSQL
al
te
rn
at
ives.
This
the
n
sug
ge
sts
a
Be
nch
m
ark
in
g a
m
on
g pee
rs
w
it
h
scenari
os
produce
d
in
web
app
li
cat
io
n
act
ivit
y used
to de
te
rm
ine the b
es
t fit
for
va
rio
us
sce
nar
i
os
.
Be
nc
hm
ark
in
g
in
this
resp
ect
re
fers
to
a
per
f
or
m
ance
eval
uatio
n
of
N
oS
Q
L
so
luti
ons
pro
po
se
d
or
i
n
us
e.
T
he
dem
a
nd
t
her
e
f
or
e
is
that
sam
ple
int
eracti
on
s
m
i
m
i
ckin
g
sim
i
la
r
be
hav
i
our
or
act
ion
s
as
case
m
ay
be
in
su
c
h
web
a
pp
li
cat
io
ns
be
us
e
d
in
a
pro
ba
bili
sti
c
or
det
erm
inistic
fashi
on
to
be
nc
hma
rk
t
he
perform
ance
of
sel
ect
ed
N
oSQL
data
base
.
On
ly
in
su
c
h
w
ay
s
can
it
s
sel
e
ct
ion
be
deem
e
d
reas
on
a
bly
su
it
abl
e
to b
e
f
ast
e
r
a
nd m
or
e su
it
ed
t
o a pa
rtic
ular
set o
f user
interac
ti
on
tha
n
a
pee
r.
NoSQL
databa
ses
can
be
cl
a
ssed
int
o
f
our
cat
egories
nam
el
y:
Key
-
value
stores,
D
ocum
ent
stores
,
W
i
de
c
olu
m
n
stores
,
G
ra
ph
Databases
[
4
]
,
[
5].
A
key
-
va
lue
sto
re
ca
n
be
viewe
d
as
a
colle
ct
ion
of
r
egiste
rs,
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
2
01
8
:
5333
-
5341
5334
each
ide
ntifie
d
by
a
key
[6
]
.
A
com
m
on
us
e
case
f
or
t
hese
syst
em
s
is
as
a
la
ye
r
i
n
the
data
-
retr
ie
val
hierar
c
hy:
a
c
ache
f
or
ex
pe
ns
ive
-
to
-
obta
in
values
,
in
de
xe
d
by
uniq
ue
keys
[
7].
[
5]
Asser
t
t
hat
ke
y
-
value
stores
a
re
ge
ne
rall
y
go
od
s
olut
ion
s
if
these
ha
ve
a
sim
ple
a
pp
li
cat
io
n
with
on
ly
one
kind
of
ob
j
ect
,
an
d
on
ly
need
t
o
lo
ok
up
ob
j
ect
s
base
d
on
on
e
at
tri
bute
.
E
xam
ples
include
Re
dis,
Dynam
o,
Me
m
cached
,
V
olde
r
m
or
t
et
c. D
oc
um
ent stor
es als
o
kn
own
as
docum
ent
-
or
ie
nte
d
dat
abases sto
re
docum
ent
-
or
ie
nted
data i
n
t
he
f
orm
o
f
Bi
nar
y
Javasc
r
ipt
Object
N
otati
on
(BS
ON)
[8
]
or
Ja
vasc
ript
O
bj
ect
Nota
ti
on
(JSON
).
These
syst
em
s
are
app
eal
in
g
to
Web
2.0
pro
gra
m
m
ers
since
these
are
ge
ne
rall
y
su
pport
ed
by
JSON
as
their
data
m
od
el
[9
]
.
Un
li
ke
the
ke
y
-
value
store
s,
these
syst
em
s
ge
ner
al
ly
suppo
rt
seco
nda
ry
ind
e
xes
a
nd
m
ulti
ple
typ
es
of
do
c
um
ents
(ob
j
ect
s)
per
data
base
a
nd
neste
d
doc
um
ents
or
li
sts
[10].
Th
ese
are
not
re
quire
d
to
a
dh
e
r
e
to
a
sta
nd
a
rd
sch
e
m
a,
the
flexibil
it
y
of
JSON
al
lows
t
he
us
er
t
o
work
w
it
h
dat
a
without
ha
vi
ng
t
o
de
fine
a schem
a
upfro
nt [8
]
,
[
9]
. E
xam
ples o
f suc
h database i
s Mo
ngoDB, C
ou
c
hDB et
c.
W
i
de
c
olu
m
n
stores
al
s
o
ref
e
rr
e
d
by
so
m
e
as
exten
sible
re
cord
st
or
es
see
m
to
hav
e
bee
n
m
otivate
d
by
Goo
gle’s
s
uccess
with
Bi
gTa
ble
[10].
A
colum
n
-
store
stores
each
a
tt
r
ibu
te
in
a
database
ta
ble
sepa
ratel
y,
su
c
h
that
su
cce
ssive
values
of
that
at
trib
ute
a
re
st
or
e
d
c
onse
cutivel
y
[
11
]
.
I
t
can
be
ar
gued
that
the
e
qu
i
va
le
nt
of
relat
ion
al
da
ta
bases
f
or
Bi
g
Data,
retai
ni
ng
th
e
noti
on
of
ta
bles,
rows
a
nd
c
olu
m
ns
[
5].
W
i
de
Colu
m
n
databases
a
re
base
d
on
a
hybr
i
d
ap
proach
that
reli
es
on
r
el
at
ion
al
datab
ases
declarat
iv
e
char
act
erist
ic
s
an
d
var
i
ou
s
k
ey
-
va
lue sto
res
sc
he
m
a [4
]
. E
xam
ples o
f
these
inc
lud
es
H
Ba
se
, C
assan
dr
a,
and
A
cc
um
ulo
etc.
Gr
a
ph
Databas
es
are
su
it
able
to
store
not
on
ly
info
rm
at
ion
about
obje
ct
s
bu
t
al
so
al
l
relat
ion
s
hip
s
that
exist
a
m
on
g
them
[4]
.
In
this re
gard,
G
raph d
at
ab
ases
em
plo
ys
Gr
a
ph
the
or
y
c
on
ce
pts
li
ke
no
des
a
nd
e
dg
e
s.
Node
s
are
entit
ie
s
in
the
data
do
m
ain
re
pr
ese
ntin
g
a
tup
le
or
r
ow
in
a
database
,
or
a
n
XM
L
el
e
m
ent
and
e
dg
e
s
are
the
relat
io
ns
hi
p
betwee
n
tw
o
entit
ie
s
li
ke
a
foreig
n
key/
pr
i
m
ary
key
relat
ion
s
hip
[
5
]
,
[
12]
.
E
xam
ples
inclu
de
Neo4J, a
nd Orie
ntDB.
Ho
wever,
t
he
f
oc
us
of this re
searc
h work i
s
on k
ey
v
al
ue
d
at
a
sto
r
es.
1.1.
In
-
Mem
ory and O
n
Disk Ke
y
-
V
alu
e
Data st
ores
Data
stores
ca
n
be
cl
assed
as
ei
ther
In
-
Me
m
or
y
or
O
n
Dis
k
data
stores.
When
in
-
m
e
m
or
y
data
stores
run,
data
is
e
nt
irel
y
loaded
i
nto
m
e
m
or
y,
s
o
al
l
it
s
operat
ion
s
a
re
run
from
m
e
m
or
y
[1
3].
Ty
pical
ly
,
su
c
h
syst
e
m
s
m
ay
r
equ
i
re
that
data
be
store
d
pe
rio
dical
ly
and
asy
nchron
ously
on
dis
k
but
al
l
wo
r
king
da
ta
is
retrieve
d
from
m
e
m
or
y.
[1
0]
Sh
ows
that
in
-
m
e
m
or
y
data
can
be
copi
ed
to
disk
f
or
bac
kup
or
syst
e
m
sh
ut
dow
n.
T
he
key
advanta
ges
of
in
-
m
e
m
or
y
data
stores
are
lo
w
la
te
ncy
and
im
pr
ove
d
thr
ough
pu
t
.
In
-
m
e
m
or
y
key
-
va
lue
stora
ge
al
so
re
qu
ire
s
low
over
head
ne
twork
com
m
un
ic
at
ion
bet
we
en
cl
ie
nts
and
serv
e
rs
[13],
t
his
c
on
tr
ibu
te
s
sig
nificantl
y
to
it
s
high
t
hroug
hput.
Re
dis
a
nd
Me
m
cached
a
re
t
ypic
al
exam
ples
of
i
n
-
Mem
or
y key
-
va
lue D
at
a
store
s [7
]
,
[
14]
.
Anothe
r
Ca
te
gory
of
key
-
val
ue
data
store
s
are
Disk
-
based
key
-
val
ue
dat
a
stores.
Ty
pic
al
ly
,
disk
-
base
d
data
sto
r
es
can
be
Distr
ibu
te
d
sto
ra
ge
syst
e
m
s
[1
5],
or
Si
ng
le
node
pl
at
fo
rm
s,
stora
ges
t
hat
hold
da
ta
i
n
hard
dis
k
with
fr
eq
ue
ntly
ac
cessed
data
in
m
e
m
or
y,
or
data
stores
that
can
store
data
in
RAM,
bu
t
it
al
so
per
m
it
s
plu
gg
i
ng
in
a
sto
rag
e
eng
ine
[10].
The
ad
va
ntage
of
on
dis
k
capab
il
it
ie
s
includes
re
du
ci
ng
c
os
t
pe
r
byte
of
st
or
a
ge
and
i
ncr
easi
ng
st
or
a
ge
ca
pa
ci
ty
.
In
dee
d,
on
dis
k
data
s
tores
ca
n
ser
ve
as
al
te
rn
at
iv
e
to
in
m
e
m
or
y
ty
pes
wh
e
n
e
xh
a
us
ti
on
of
m
e
m
or
y
sp
ace
is
antic
ipate
d
or
e
xpec
te
d
[
15
]
.
Also
,
these
are
ref
e
r
red
t
o
as p
e
rsiste
nt
ke
y
-
value
sto
res whic
h
incl
ude
data sto
res
s
uc
h
as
Ber
keley
DB, Vol
dem
or
t and Ri
ak [1
6]
.
1.2.
Redis
Re
dis
is
an
ope
n
sou
rce
(BS
D
li
censed
)
,
in
-
m
e
m
or
y
data
structu
re
sto
re,
us
e
d
as
data
ba
se,
cache
a
nd
m
essage b
r
oke
r
[14]. Th
e
data
m
od
el
is k
ey
-
val
ue,
alt
houg
h
m
any d
iffer
e
nt k
in
ds
of d
at
a ty
pes
are su
pport
ed:
Strin
gs
,
List
s,
Sets,
Sorte
d
S
et
s,
Hash
e
s,
H
yperL
ogLo
gs
,
and
Bi
tm
aps
[
17
]
.
Re
dis
has
bu
il
t
-
in
rep
li
ca
ti
on
,
it
can
be
re
plica
te
d
us
in
g
the
m
ast
er
-
sla
ve
m
od
el
and
a
m
ast
er
can
ha
ve
m
ulti
ple
sl
aves
[
14
]
,
[
3].
Re
dis
pro
vid
es
acce
s
s
to
m
utable
da
ta
structur
e
s
vi
a
a
set
of
com
m
and
s,
wh
ic
h
are
sent
us
i
ng
a
serv
e
r
-
cl
ie
nt
m
od
el
with
TCP
so
c
ke
ts
and
a
si
m
ple
protoc
ol
[17].
Re
dis
can
be
us
e
d
as
a
Least
recently
us
ed
(LRU
)
cache,
us
in
g
an
a
ppr
ox
im
ate
LRU
al
gorith
m
to
evict
old
data
as
a
ne
w
on
e
is
a
dd
e
d
[14].
It
al
s
o
offers
scri
pting
cap
abili
ty
us
in
g
L
ua,
a
powe
rful,
li
gh
t
-
wei
gh
t
scri
pti
ng
la
ngua
ge
wr
it
te
n
in
C
a
nd
em
bedded
in
Re
dis
[
18
]
.
Re
dis
su
pp
or
ts
a
utom
at
ic
fail
ov
er
if
a
m
ast
er
is
no
t
w
orkin
g
as
exp
ect
e
d
us
in
g
a
featu
re
cal
le
d
Re
dis
Sentin
el
,
this
sta
rts
a
fail
ov
e
r
process
wh
e
r
e
a
sla
ve
is
pr
om
oted
to
m
as
te
r,
ad
diti
on
al
sla
ves
are
a
lso
reconfigu
red
t
o
us
e
the
ne
w
m
ast
e
r
[
14]
.
S
hardin
g
is
e
xecu
te
d
via
Re
dis
cl
us
te
r
a
platfo
rm
wh
e
re
data
is
autom
at
ic
ally
sh
ar
de
d
acro
s
s m
ulti
ple
Redis
nodes
[14
]
.
1.3.
SSD
B
SSD
B i
s a f
ast
NoSQL data
ba
se f
or sto
rin
g
bi
g
li
st of
b
il
li
ons of elem
ents;
it
su
pport
s d
at
a stru
ct
ures
includi
ng
Key
-
Value
pai
r,
List
,
Ma
p
or
Ha
sh
an
d
Sorte
d
Set
[1
9].
S
SDB
is
wr
it
te
n
in
C/
C+
+
with
Goo
gle
LevelDB
as
it
s
storag
e
e
ng
i
ne
[2
0].
Co
ncei
ved
by
i
ts
autho
r
as
an
al
te
r
na
ti
ve
to
Re
dis
[19,
20]
,
it
su
pport
s
Re
dis
netw
ork
protoc
ol
and
op
e
n
-
s
ource
d
Re
dis
cl
ie
nts
[21].
SSD
B
oth
e
r
featu
res
include
Re
pli
cat
ion
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
Perf
orma
nce B
enchm
ar
ki
ng
of
Key
-
Val
ue Store N
oSQL D
at
abas
es
(
O
m
oruyi
Osemwe
gie
)
5335
(Mast
er
-
Slave
and
Mast
er
-
Ma
ste
r
co
nf
ig
urat
ion), Tim
e to Li
ve
key ex
pirati
on
(can
se
r
ve
a
s a p
ersi
ste
nt c
ache
serv
ic
e
)
[
21]
a
nd easy
to
u
se
cl
ie
nt A
P
Is for
Dev
el
op
m
ent a
nd D
e
plo
ym
ent.
2.
RESEA
R
CH MET
HO
D
2.
1.
Benchma
rking
To
ol
an
d System
S
pe
c
ifica
tio
n
Be
nch
m
ark
in
g
of
N
oSQL
Databases
wit
h
a
ny
sta
ndar
d
ben
c
hm
ark
too
ls
is
done
us
in
g
t
w
o
appr
oach
es
na
m
el
y:
trace
and
vect
or
-
base
d
load
(
databas
e
op
e
rati
ons)
gen
e
rati
on
[
22
]
,
[
23
]
.
T
race
base
d
ben
c
hm
ark
sys
tem
s
us
e
act
ua
l
app
li
cat
ion
work
l
oa
d
ge
ne
rated
from
sp
eci
fic
app
li
cat
io
ns
ov
e
rtim
e.
Wh
il
st
Vecto
r
base
d
be
nch
m
ark
syst
e
m
s
create
s
app
li
cat
ion
be
ha
viours
us
in
g
ve
ct
or
s
a
nd
a
pp
l
yi
ng
the
vect
ors
us
i
ng
known
sta
ti
sti
cal
distrib
utio
n
m
od
el
s,
m
i
m
ic
kin
g
act
ual
ap
plica
ti
on
re
qu
e
st
an
d
res
ponse
in
ha
r
dware
or
virtu
al
platf
orm
.
Be
nch
m
ark
too
ls
can
be
c
la
ssifie
d
as
ei
ther
in
buil
t
or
custom
.
An
exa
m
ple
of
the
f
orm
er
is
red
is
-
be
nc
hm
a
rk.
Cu
stom
ben
chm
ark
t
oo
l
s
inclu
de
Ya
hoo!
Cl
oud
S
yst
e
m
Be
nchm
ark
(
YCSB
)
[
24
]
,
Bi
gBench
[25
]
and
Gray
Sort
[26].
T
he
ob
j
e
ct
ive
of
t
his
be
nch
m
ark
in
g
process
is
to
c
ompa
re
t
he
pe
rform
ance
of
Re
dis
an
d
S
SD
B
N
oSQL
databases
us
in
g
sing
le
no
de
instances
.
The
resu
lt
s
w
ou
l
d
validat
e
the
cl
aim
s
of
SSD
B
’s
s
uitab
il
ity
as
an
al
te
r
native
to
Re
dis
as
the
a
uthor
s
hav
e
sug
gested
[19]
.
T
he
t
oo
l
of
c
hoic
e
is
Y
CSB
,
YCSB
’s
pl
ug
i
n
-
base
d
a
rc
hite
ct
ur
e
a
nd
ease
of
exte
ns
ibil
it
y
us
i
ng
scri
pts
[
27
]
m
akes
it
a
sp
le
ndid
ch
oic
e.
[
4]
detai
ls YCSB’
s u
se
in
m
easuri
ng
t
he per
for
m
ance of fo
ur
NoSQL
syst
em
s inclu
ding Re
dis. [2]
Descr
i
bes
tw
o
YCSB
be
nch
m
ark
ti
ers
nam
ely:
Perf
or
m
ance
and
Scal
in
g.
The
f
ocu
s
f
or
this
stud
y
is
to
Be
nch
m
ark
SS
DB’s
perform
ance
in
com
par
is
on
to
Re
dis
f
or
a
range
of
Wor
klo
a
ds
.
These
work
l
oa
ds
im
i
ta
te
a
var
ie
ty
of
we
b
app
li
cat
io
n
re
quest
beh
a
viour
s li
ke heavy
re
ad
a
nd writ
e sc
enar
i
os
. T
he
workloa
ds
c
onsi
der
e
d
i
nclu
des:
Work
l
oad
A
(
Heavy
U
pdati
ng
)
In this
wor
klo
a
d 50%
of
t
he o
per
at
io
ns
a
re
r
eads a
nd
50
%
are
wr
it
es .
Work
l
oad
B
(Heavy Re
ad)
In this
wor
klo
a
d 95%
of
t
he o
per
at
io
ns
a
re
r
eads a
nd the
r
e
st 5% a
re
wr
it
e
s.
Work
l
oad
C
(Only Re
ad)
Wor
klo
a
d wit
h 1
00% r
ea
d o
pe
rati
on
s
.
Work
l
oad D
(
Re
ad La
te
st)
Wor
klo
a
d wit
h 9
5% read
Ope
rati
on
s
and
5%
inser
t
operati
ons. Wo
r
klo
a
d
i
ns
erts
n
e
w rec
ords
a
nd the
m
os
t
recently
inse
rted reco
rds are
the m
os
t p
opula
r.
Work
l
oad
E
Wor
klo
a
d wit
h 9
5%
S
hort
ra
nges S
can
Oper
at
ion
s a
nd
5% i
ns
ert
op
e
rati
on
s. Wo
r
klo
a
d q
uer
ie
s
short
ra
ng
e
s
of r
ec
ords, inst
ead
of in
div
id
ua
l reco
rds.
Work
l
oad
F
Wor
klo
a
d wh
e
re th
e
cli
ent w
i
ll
r
ead a
r
ec
ord
, m
od
ify
it
, and
w
rite
bac
k
the
ch
a
ng
e
s.
Ta
bl
e 1
show
s the
detai
ls sy
stem
co
nfi
gurati
on
a
nd spec
ific
at
io
n for the
b
e
nc
hm
ark
in
g proce
ss.m
anu
script.
Table
1: D
et
ai
li
ng
Syst
em
co
nf
i
gurati
on a
nd Sp
eci
ficat
ion
Proces
so
r
Intel Pen
tiu
m
C
PU
B9
6
0
Clo
ck
Speed
2
.20
GHz
Nu
m
b
e
r
o
f
Co
res
2
Nu
m
b
e
r
o
f
T
h
read
s
2
Ho
st Ins
tructio
n
Set
6
4
bit
Ho
st Op
erating
Sys
te
m
W
in
d
o
ws
Ho
st Me
m
o
ry
4
0
9
6
MB
Virtual Op
e
rating
Sy
ste
m
Cen
tOS 6.3
Virtual
Me
m
o
r
y
7
5
6
MB
Virtual Ins
tructio
n
Set
3
2
bit
Kernel
Linu
x
2.6
.32
-
2
7
9
.e
l6
.i68
6
Jav
a
Jav
a SE
Ru
n
ti
m
e
1.8
.0_
4
5
YCSB
Ver
sio
n
0
.1.4
Red
is Versio
n
3
.2.0
SSDB Ve
rsio
n
1
.9.3
Virtual
Machin
e
Orac
le
VirtualB
o
x
3.
RESU
LT
S
A
ND AN
ALYSIS
3.1.
Workl
oad
A (
Hea
vy Upd
at
i
ng
)
The
resu
lt
s
of
the
Hea
vy
Update
ope
rati
ons
are
s
how
n
i
n
Fig
ur
es
1,
Fig
ur
e
2
a
nd
Fi
gure
3.
In
this
resu
lt
,
SS
DB
outpe
rfor
m
s
Red
is
as
the
nu
m
ber
of
th
read
s
increases
a
nd
i
ts
through
pu
t
div
e
rg
es
si
gn
if
ic
antly
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
2
01
8
:
5333
-
5341
5336
with
an
i
ncr
eas
e
in
the
num
ber
of
threa
ds.
T
his
sug
gests
th
at
an
increase
i
n
softwa
re
th
re
ads
will
increa
se
the
perform
ance
of SS
DB as s
ho
wn in t
he
Fi
gur
es 1
-
3
Figure
1.
Com
par
is
on of Re
di
s and SSDB
w
it
h
1000
Re
cords
Figure
2.
Com
par
is
on of Re
di
s and SSDB
w
it
h
5000
Re
cords
Figure
3.
Com
par
is
on of Re
di
s and SSDB
w
it
h
1000
0
Re
c
ords
and
1000
O
per
at
io
ns
3.2.
Workl
oad
B
(
Hea
vy R
e
ad)
The
res
ult
of
the
Hea
vy
Re
ad
w
orkloa
d
as
sh
ow
n
in
Fig
ure
4,
Fi
gure
5
and
Fi
gure
6
ha
s
si
m
i
la
rity
with
the
heav
y
update
ope
rati
on
s
.
SS
DB
ou
t
perform
s
Re
dis
for
Hea
vy
Re
ad
ope
rati
on
s
a
s
show
n
in
Fig
ur
e
1,
Figure
2
an
d
Figure
3.
H
ow
ever
this
a
dva
ntage
of
perf
orm
ance
seem
s
to
be
lo
st
as
t
he
am
ount
of
record
s
appr
oach
es
10,000.
Redis cle
a
rly
seem
s to
reco
ve
r
a
ny lost
gro
unds
at t
his
le
vel.
Figure
4
.
Com
par
is
on of Re
di
s and SSDB
w
it
h
1000
Re
cords
a
nd
1000
O
per
at
io
ns
Figure
5
.
Com
par
is
on of Re
di
s and SSDB
w
it
h
5000
Re
cords
a
nd
1000
O
per
at
io
ns
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
Perf
orma
nce B
enchm
ar
ki
ng
of
Key
-
Val
ue Store N
oSQL D
at
abas
es
(
O
m
oruyi
Osemwe
gie
)
5337
Figure
6
.
Com
par
is
on of Re
di
s and
US
S
DB
with
10000 R
e
cords
a
nd
1000
Operati
ons
3.3.
Workl
oad
C (
Only Re
ad)
The
resu
lt
of
the
thr
ough
pu
t
for
S
SD
B
as
show
n
i
n
Fig
ur
es
7,
Fig
ure
8
a
nd
Fig
ur
e
9
div
e
rge
s
sign
ific
a
ntly
ag
ai
ns
t t
hat of R
edis as t
he num
ber
o
f
t
hr
ea
ds de
plo
ye
d
inc
r
eases
.
Figure
7
.
Com
par
is
on of Re
di
s and SSDB
w
it
h
1000
Re
cords
a
nd
1000
O
per
at
io
ns
Figure
8
.
com
par
iso
n of Re
dis
and SSDB
wit
h
5000 Rec
ords and 1
000 O
pe
r
at
ion
s
Figure
9
.
Com
par
is
on of Re
di
s and SSDB
w
it
h
1000
0
Re
c
ords
and
1000
O
per
at
io
ns
3.4.
Workl
oad
D (
Read La
tes
t)
The
Fig
ure
10,
Figure
11
an
d
Figure
12
s
hows,
how
the
re
su
lt
of
S
SD
B
t
hro
ughput
sur
pa
sses
that
of
Re
dis.
T
his als
o
s
hows
a sig
ni
ficant
boos
t
of
the th
rou
ghput
as the
n
um
ber
o
f
thr
ea
ds i
ncrea
ses to 8.
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
2
01
8
:
5333
-
5341
5338
Figure
10
.
C
om
par
ison
of R
edis a
nd SSDB
w
it
h
1000 Rec
ords and 1
000 O
pe
r
at
ion
s
Figure
11
.
C
om
par
ison
of Redis a
nd SSDB
w
it
h
5000 Rec
ords and 1
000 O
pe
r
at
ion
s
Figure
12
.
C
om
par
ison
of Redis a
nd USSD
B wit
h 1
0000
Re
cords
a
nd
1000
O
per
at
io
ns
3.5.
Workl
oad E
The
Fig
ur
e
13,
Fig
ur
e
14
a
nd Figure 15
s
ho
ws,
ho
w
the
R
edis
cl
early
o
ut
perform
s
SSDB
in
te
rm
s
of
thr
oughput. T
hi
s ind
ic
at
es t
ha
t t
he
SS
DB is
un
s
uitable
for SC
AN ope
rati
on
s
.
Figure
13
.
C
om
par
ison
of Redis a
nd SSDB
w
it
h
1000 Rec
ords and 1
0
00 Ope
r
at
ion
s
Figure
14
.
C
om
par
ison
of Redis a
nd SSDB
w
it
h
5000 Rec
ords and 1
000 O
pe
r
at
ion
s
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
Perf
orma
nce B
enchm
ar
ki
ng
of
Key
-
Val
ue Store N
oSQL D
at
abas
es
(
O
m
oruyi
Osemwe
gie
)
5339
Figure
15
.
C
om
par
ison
of Redis a
nd SSDB
w
it
h 1
0000 Re
cords
a
nd
1000
Operati
ons
3.6.
Workl
oad F
In
the
Re
ad
-
M
od
i
fy
-
Wr
it
e
Wo
r
klo
a
d
as
sho
wn
in
Fig
ure
16,
Fig
ur
e
17
and
Fi
gure
18,
there
are
sh
a
des
of
sim
ilarity
to
Work
l
oad
D
(Rea
d
L
at
est
)
see
Figure
10
to
Fig
ure
12.
SS
DB
pac
es
Re
dis
for
outp
uts
from
1
2
,
and
4 t
hr
ea
ds
res
pec
ti
vely
. H
owe
ve
r
the
re is a
d
i
ve
rg
e
nce
w
hen thr
ea
d
le
ng
t
h
in
creases t
o 8.
Figure
16
.
C
om
par
ison
of Redis a
nd SSDB
w
it
h
1000 Rec
ords and 1
000 O
pe
r
at
ion
s
Figure
17
.
C
om
par
ison
of Redis a
nd SSDB
with
5000 Rec
ords
a
nd
1000
Op
e
rati
ons
Figure
18
.
C
om
par
ison
of Redis a
nd SSDB
w
it
h 1
0000 Re
cords
a
nd
1000
Operati
ons
F
or
t
he
series
of
e
xperim
ents
cond
ucted,
on
ly
sing
le
node
instances
of
both
databases
wer
e
us
e
d.
The
te
st
carrie
d
out
involve
d
al
l
six
(6
)
g
e
ne
ric
work
l
oads
of
YCSB
(
W
orkloa
d
A
to
Wor
k
loa
d
F)
.
SSD
B
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
2
01
8
:
5333
-
5341
5340
perform
s
bet
te
r
tha
n
Re
dis’s
thr
ough
pu
t
in
hea
vy
r
ead
operati
ons
by
a
sig
nificant
m
arg
i
n.
H
ow
e
ve
r,
t
he
m
arg
in
in
dicat
es
a
sign
i
fican
t
reducti
on
with
rea
d
-
m
od
ify
and
w
rite
op
e
r
at
ion
s.
T
his
s
hows
t
hat
de
vel
op
e
rs
can
co
ns
i
der
a
doptin
g
SS
DB
for
el
ast
ic
it
y
in
app
li
cat
ion
sc
enar
i
os
w
he
re
updates,
hea
vy
read
&
rea
d
-
m
od
ify
and
wr
it
e
oper
at
ion
s
are
unde
rtake
n.
Wh
il
e
,
SSD
B
’s
cl
ai
m
to
be
a
su
it
able
al
te
rn
at
iv
e
to
Re
dis
has
been
sh
ow
n
to
be
va
li
d
for
F
ive
out
of
si
x
w
ork
loads
,
it
is
no
t
su
it
ed
f
or
s
ho
rt
or
sm
al
l
(<
10,00
0)
range
scan
qu
e
ries.
4.
CONCL
US
I
O
N
In
c
oncl
us
i
on,
the
phe
no
m
eno
m
s
that
has
le
d
to
the
i
ncr
ea
sed
visibil
it
y
of
N
oSQL
is
th
e
risin
g
nee
d
for
un
st
ru
ct
ur
e
d
data
an
d
do
c
um
ents
in
Mobi
le
com
pu
ti
ng
and
S
ocial
net
work
we
bs
it
es
[28]
.
Wh
il
st
c
heap
e
r
and
faster
m
em
or
y
un
it
s
are
no
t
ru
le
d
out,
the
gro
wing
t
rend
of
Bi
g
D
at
a
and
t
he
I
nt
ern
et
of
T
hings
is
decen
t
rali
zed
database
syst
em
s
that
i
m
pr
ov
e
fau
lt
tole
ra
nce
in
data
bas
e
syst
e
m
s.
Per
form
ance
conc
ern
s
are
no
t
only
influ
e
nced
by
locat
ion
al
one
b
ut
by
data
secur
it
y
issues
al
so
.
Se
cur
it
y
con
ce
r
ns
arise
because
of
the
natu
re
an
d
c
ha
racteri
sti
cs
of
Bi
g
Data
(the
huge
vo
l
um
e,
velocit
y,
va
riet
y
and
ver
aci
ty
of
data)
[
29]
.
On
e
of
su
c
h
co
ncerns
is
ho
w
to
qu
e
r
y
encr
ypte
d
da
ta
base
syst
e
m
s
without
degra
ding
pe
rfor
m
ance
of
ap
plica
ti
on
s
[30,
3
1]. A
lt
ho
ugh
this
st
ud
y has
not
ad
dress
ed
su
c
h
sec
ur
it
y
issues
it
is
ce
rtai
n
that
this will
featur
e
in lots
of
stud
ie
s
goin
g f
orward.
ACKN
OWLE
DGE
MENTS
We ac
knowle
dge the
s
upport
of Co
ven
a
nt
U
niv
e
rsity
in
c
onduct
in
g
this
r
esearch
and t
he
co
st
of pu
blica
ti
on
.
REFERE
NCE
S
[1]
C.
U.
Kum
ara
sin
ghe,
K
.
L.
D
.
U. L
i
y
ana
g
e, W.A.
T
.
Madushanka
an
d
R.
A.C
.
L. Men
dis.
(2015
,
Sept
e
m
ber
).
Perform
anc
e
Co
m
par
ison o
f
No
SQ
L
Data
base
s
i
n
Ps
eudo
Distrib
ute
d
Mode
:
C
assandra
,
MongoD
B
&
R
edi
s
[Onlin
e]
.
Avai
lable:
htt
ps://
ww
w.re
se
arc
hga
te.ne
t
/prof
il
e/Tirosha
n_Ma
dushanka
/publ
i
c
at
ion/
281629653
_Perform
anc
e_C
om
par
ison_of
_NoS
QL_Dat
aba
ses_in_Ps
eudo_Distri
bute
d_M
ode_Ca
ss
andr
a_
MongoD
B_Redi
s/li
nks/55f113ba
08ae
de
cb68f
fd2
9
4.
pdf, Access
ed: Jul
y
.
1
,
2016
[2]
B.
F.
Cooper,
A
.
Silbe
rst
ei
n
,
E.
Ta
m
,
R.
R
amakri
shnan,
and
R.
S
ea
rs,
“
Benc
hm
ar
king
cl
oud
serv
i
ng
s
y
stems
with
y
csb
”
.
In
Proceedi
ngs
of
th
e
1st
ACM
sympos
ium
on
Cloud
computi
ng
(New
York,
NY
,
US
A,
2010),
SoC
C
’10,
ACM
,
pp.
143
-
1
54.
[3]
Y.
Abubaka
r
,
T
.
S.
Ad
e
y
i,
an
d
I.
G.
Aut
a,
"P
erf
orm
anc
e
Eva
lu
at
ion
of
N
oSQ
L
S
y
stems
using
YCS
B
in
a
Resourc
e
Aus
te
r
e
Envi
ronm
ent,"
Inte
rnational
J
ournal
of
Appl
i
ed
Information
Syste
ms
,
vol
.
7,
pp.
23
-
27,
Sep
.
2014.
[4]
[A.
Ous
sous
,
F.
Z.
Benjel
loun,
A.A.
La
h
c
en,
S
.
Bel
fki
h,
"Com
par
ison
and
Cla
s
sific
a
ti
on
of
NoS
QL
Data
base
s
for
Big
Data,"
in
Pr
oc.
o
f
th
e
2015
I
nte
rnational
Co
nfe
renc
e
on
Bi
g
Data,
Cloud
and
Appl
i
cat
ions
,
B
DCA
2015,
25
-
2
6
Ma
y
2015,
Tetu
an,
Morocc
o
[O
nli
ne]
.
Availabl
e
:
Resea
r
chGa
te
,
htt
ps://
ww
w.
res
ea
rch
g
ate.
ne
t.
[
Acc
essed:
24
June
2016]
.
[5]
H.
Khaz
ae
i
,
M.
Fokaef
s,
S.
Zare
ia
n,
N.
B
ei
gi
-
M
ohamm
adi
,
B.
R
amprasad,
M.Sh
te
rn,
P.
Ga
ikwa
d,
and
M.
Litoiu
.
“
How
do
I
choose
the
right
NoS
QL
soluti
on
?
A
c
om
pre
hensive
th
eor
etical
and
ex
per
imental
surve
y
.
”
In
:
Subm
i
tte
d
to
Journal
of
Bi
g
Data
and
Information
Anal
y
tic
s(
BDIA
)
(2015).
[Online
]
.
Availabl
e:
htt
ps://
ww
w.re
se
arc
hga
te.ne
t
/prof
il
e/
H
amze
h_Kha
za
e
i/
publ
ic
a
ti
on/
282679529_How
_Do_I_Choos
e
_The
_Right
_N
osql_S
olut
ion_A_Com
pre
hensive_The
ore
ti
c
al
_An
d_Expe
riment
al_S
urve
y
/
li
n
ks/5
618781808ae
04
4edba
d2437.
pdf
.
Acc
essed:Jun. 2
9,
2016
.
[6]
E.
Anderson,
X
.
Li
,
M
.
A.
Shah
,
J.
Tucek,
and
J.
J.
W
y
lie,
“
W
hat
consiste
n
c
y
do
es
y
our
k
e
y
-
va
lu
e
store
ac
tu
al
l
y
provide?
”
HotD
ep
,
vo
l. 10, pp. 1
–
16,
2010
.
[7]
B.
Atikogl
u,
Y. X
u,
E.
Frac
ht
en
ber
g,
S.
Jian
g
,
a
nd
M.
Pale
cz
n
y
.
W
orkloa
d
ana
l
y
s
is of
a
la
rge
-
sc
ale
ke
y
-
val
u
e
stor
e.
In
Proceedi
ngs
of
th
e
SIGMETR
ICS’12
,
June
20
12.
[8]
K.
Ma,
A.
Abra
ham,
“
Towa
rd
l
i
ghtwei
ght
tra
nsp
are
nt
data
m
iddleware
in
suppor
t
of
do
cument
st
ore
s”
in
W
ICT
2013:
Proc
ee
din
gs of
th
e thi
rd
W
orld Congre
ss
on
Information
an
d
Comm
unic
ati
o
n
Technol
og
ie
s
(
2013)
[9]
C.
Chasseur,
Y.
Li
,
and
J.
M.
Pa
te
l
.
En
abl
ing
JS
ON
document
stores
in
r
el
a
ti
ona
l
s
y
st
ems
.
In
Pr
oce
ed
ings
of
th
e
16th
Int
ernati
on
al
Workshop on
the
W
eb
and
Dat
abases (
We
bDB),
pag
es
1
-
6
,
20
1
3.
[10]
R.
C
at
t
ell,
“
Scalable sql and
nos
ql
da
ta store
s,
”
ACM
SIGMO
D
Re
cord
,
vol
.
39
,
no.
4
,
pp
.
12
–
27
,
2010.
[11]
D.J.
Abad
i. Col
um
n
stores
for
w
ide
and
sparse
d
at
a
.
In
CIDR,
As
il
om
ar,
CA
,
US
A,
2007.
[12]
V.
Kac
holi
a
,
S.
Pandit
,
S.
Chakr
aba
rt
i,
S.
Suda
rshan,
R
.
Desa
i
,
a
nd
H.
Kara
m
belkar,
"Bid
ire
c
ti
on
al
exp
ansion
fo
r
ke
y
word
sea
r
ch on gr
aph
d
ataba
s
es,
" i
n
Proc
.
of
VLDB
Confe
ren
ce
,
2005,
pp.
50
5
-
516.
[13]
J.
Han,
E.
Haih
ong,
G.
Le
,
and
J.
Du,
"
Surve
y
on
No
SQ
L
dat
aba
se,
"
In
Pe
rv
asive
Computin
g
and
Appl
ic
ations
(
ICPCA
)
,
2011
6th
Int
ernati
onal
Confe
renc
e
on
,
Oct.
2011
,
pp
.
3
63
-
366.
[14]
[Online
]
.
Availab
le
:
htt
p
:/
/r
edi
s.io
Acc
essed:
Jul
y
.
1,
2016
.
[15]
Online
]
.
Availabl
e:
htt
p
:/
/www
.
id
e
awu.
com/blog/
p
ost/c
ategor
y
/ssd
b,
Ac
ce
ss
ed:
Jul
y
.
1
,
2016
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
Perf
orma
nce B
enchm
ar
ki
ng
of
Key
-
Val
ue Store N
oSQL D
at
abas
es
(
O
m
oruyi
Osemwe
gie
)
5341
[16]
Kata
rina
Grolin
ger
,
W
il
son
A.
Higashino,
Abhi
na
v
Ti
wari
and
Miria
m
A.
M.
C
apr
etz,
“
Data
m
ana
gement
in
cl
o
ud
envi
ronm
ent
s:
NoS
QL
and
N
ewSQ
L
dat
a
st
ore
s”,
Journal
of
Cloud
Com
puti
ng:
Advanc
es,
S
y
stems
an
d
Applic
a
ti
ons 20
13,
2:22
;
h
tt
p://
ww
w.j
ourna
lofc
l
oudcomputing.
c
om
/c
onte
nt/2/1/
22
[17]
[Online
]
.
Availab
le
:ht
tps:/
/g
it
hub
.
com/ant
ir
ez
/r
edis
Acc
essed:
Jul
y.
1
,
2016
.
[18]
R.
Ier
usal
ims
chy
,
L
.
H
.
de
Figue
ire
do,
W
.
C
el
es
.
Lua
5
.
1
R
efe
r
en
ce
Manu
al. Lua
.
org,
2006
.
[19]
[Online
]
.
Avai
lable: ht
tps:
//
ss
db.
io
Acc
essed:
Jul
y
.
2
,
2016
.
[20]
[Online
]
.
Availab
le
:https
:/
/g
it
hub
.
com/ide
awu/ss
d
b
Acc
essed:
Jul
y.
2
,
2016
.
[21]
[Online
]
.
Availab
le
:http:/
/www
.
id
ea
wu.c
om
/bl
og
/
post/c
ategor
y
/ss
db
Acc
essed:
Jul
y
.
2
,
2016
.
[22]
R.
Sear
s,
C.
Van
Inge
n
,
and
J
.
Gra
y
,
“
To
BL
OB
or
not
to
B
LOB:
La
rg
e
obj
ec
t
stor
age
in
a
dat
ab
ase
or
a
fil
es
y
s
te
m
?
,
” ar
Xiv
Prepr. c
s/07
011
68,
pp
.
1
–
11
,
2007.
[23]
M.
Selt
z
er,
D.
K
rinsk
y
,
K.
Sm
it
h
,
and
X
.
Zh
ang,
“
The
ca
se
for
ap
pli
c
at
ion
-
spe
ci
fi
c
benchm
ark
ing,”
in
Hot
Topi
cs
in
Operating
Sys
te
ms
,
1999
.
Pro
c
ee
dings o
f
th
e
S
e
ve
nth
Workshop
on
,
1999
,
pp
.
10
2
–
107.
[24]
B.
Cooper
,
YCS
B:
Yahoo!
C
lou
d
Serving
B
enc
h
m
ark
.
2017.
[25]
T.
R
abl e
t
a
l.,
“
BigBe
nch
Speci
f
i
ca
t
ion
V0.
1,
”
in
Specifying Big D
ata
Be
n
chmarks
,
Springer
,
201
4,
pp
.
164
–
201
.
[26]
“
Sort
Benc
hm
ar
k
Hom
e
Page.”
[
Online
]
.
Avai
la
b
le
:
htt
p
:/
/sortb
en
chmark.
org./.
[Acc
essed:
01
-
Jan
-
2018]
.
[27]
H.
Khaz
a
ei
et
a
l.
,
“
How
do
I
c
hoose
the
r
ight
nosql
soluti
on
?
a
compreh
ensi
ve
the
or
etical
a
nd
expe
rimen
ta
l
surve
y
.
”
[28]
S.
-
H.
Jung,
J.
-
C
.
Kim
,
and
C.
-
B.
Sim
,
“
Predic
ti
on
Data
Proce
ss
ing
Scheme
using
a
n
Artificial
Neur
al
Ne
twork
and
Data
Cl
usteri
ng
for
Big
Da
ta
,
”
I
nte
rnational
Jou
rnal
of
E
lectric
a
l
and
Computer
Engi
ne
ering
(
IJE
CE)
.
,
vol
.
6
,
n
o
.
1,
p
.
330
,
2016
.
[29]
S.
A.
Th
ane
ka
r,
K.
Subrahm
an
yam,
and
A.
B.
Bagwa
n,
“
Big
d
at
a
and
MapR
ed
uce
cha
l
le
nges
,
opportuni
ties
an
d
tre
nds,”
In
te
rnat
ional
Journal
of
El
ectric
al
and
Computer
Engi
n
ee
ring
(
IJE
CE)
.
,
vol.
6,
no.
6
,
p
p.
2911
–
2919,
2016.
[30]
Y.
D.
Jang
and
J.
H.
Kim
,
“
A
compari
son
of
the
quer
y
e
xec
ut
ion
a
lgori
t
hm
s
in
sec
ure
dat
ab
ase
s
y
s
te
m
,
”
Inte
rnational
Jo
urnal
of El
e
ct
ri
c
al
and
Comput
er
Engi
n
ee
ring
(
IJE
CE)
,
vol
.
6
,
no
.
1,
pp.
337
–
343,
2016.
[31]
Chin
onso,
Okere
ke,
Os
emw
egi
e
Om
oruy
i
,
Kenne
d
y
Okokpujie,
a
nd
Sam
uel
John.
"D
eve
lopment
of
an
Enc
r
y
pti
ng
S
y
stem
for
an
Im
age
Viewe
r
base
d
on
Hill
Ciphe
r
Algorit
hm
"
,
Cove
nant
Journ
al
Of
Engi
nee
ri
ng
Technol
ogy
1,
no.
2
(2017).
Evaluation Warning : The document was created with Spire.PDF for Python.