Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
1
3
,
No.
2
,
Febr
uar
y
201
9
, pp.
7
29
~
7
3
6
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
3
.i
2
.pp
7
29
-
7
3
6
729
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
A noble
approa
ch
to
d
evel
op dynam
ically s
ca
l
able n
amen
ode in
hadoop d
istributed file
sy
stem usi
ng secon
dary sto
rage
Tumpa
Rani
Shaha
1
,
Md.
Na
sim
Akht
ar
2
, Fatem
a
Tu
j Johor
a
3
,
Md.
Z
ak
ir
Ho
ss
ain
4
,
Most
af
ij
ur
Rahman
5
, R. B.
Ah
m
ad
6
1,
2,4
Depa
rtment
o
f
Com
pute
r
Sci
e
nce
and Engi
ne
e
ring,
Dhak
a
Uni
ver
sit
y
of
Eng
in
ee
ring
and
Tech
nolog
y
(DU
ET
)
,
Gaz
ipur, Ba
ng
ladesh
3
Instit
ute of Info
rm
at
ion
T
ec
hno
l
og
y
,
Jaha
ng
irnagar
Univer
si
t
y
(J
U),
Banglade
sh
5
D
epa
rtment of
Software
Eng
ineeri
ng,
Daffodi
l
I
nte
rna
ti
ona
l
Uni
ver
sit
y
(DIU
), Dhaka
,
Bang
la
des
h
1
Depa
rtment of
Com
pute
r
Scie
n
ce
and Engi
ne
ering,
Daffod
il
In
ternat
ion
al
Univ
er
sit
y
(DIU
),
Dhak
a,
B
anglade
sh
6
Facul
t
y
of
Infor
m
at
ic
s a
nd
Com
puti
ng,
Univer
sit
y
Sul
t
an
Zaina
l A
bidi
n
(UniSZA
),
22200
Besut,
Te
ren
gg
anu, Ma
lay
s
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
N
ov
12
, 201
8
Re
vised
Dec
13
,
2018
Accepte
d
Dec
27
, 201
8
For
sca
la
bl
e
da
t
a
storage,
Hado
op
is
widely
us
ed
nowada
y
s.
I
t
provide
s
a
distri
bute
d
fi
le
sy
stem
tha
t
store
s
dat
a
on
the
compute
nodes.
Basic
a
lly
,
i
t
rep
rese
nts
a
m
a
ster/
slav
e
arc
h
itect
ur
e
that
con
sists
of
a
Name
Node
and
copi
ous
Dat
a
N
odes.
Dat
a
Nod
es
contain
appli
ca
t
ion
da
ta
and
m
et
ada
t
a
of
appl
i
ca
t
ion
data
reside
s
in
the
Main
Mem
or
y
of
Nam
eNode
.
In
ca
ch
ed
appr
oac
h
,
they
f
rag
m
ent
the
m
etada
t
a
depe
nd
ing
on
the
la
st
acces
s
ti
m
e
and
m
ove
the
least
f
req
uentl
y
used
dat
a
to
s
ec
ondar
y
m
emor
y
.
If
th
e
req
u
este
d
dat
a
is
not
foun
d
in
m
ai
n
m
em
or
y
th
e
n
the
se
c
ondar
y
d
ata
will
be
loa
de
d
aga
in
on
the
RAM
.
So
when
the
sec
ondar
y
d
at
a
re
loa
ds
to
the
prima
r
y
m
emory
th
en
the
Nam
eNode
m
ai
n
m
emory
li
m
it
a
ti
on
ari
ses
again.
The
foc
us
of
thi
s
res
ea
rch
i
s
to
red
u
ce
the
n
amespac
e
prob
lem
of
m
ai
n
m
emor
y
and
t
o
m
ake
t
he
s
y
st
e
m
d
y
namic
al
l
y
sca
l
abl
e
.
A
n
ew
Meta
d
at
a
Fragm
ent
ation
Algorit
hm
is
proposed
tha
t
sepa
ra
te
s
th
e
m
et
ad
at
a
li
st
o
f
Nam
eNode
d
y
nami
ca
l
l
y
.
Th
e
Nam
eNode
cr
ea
t
es
Seconda
r
y
Mem
ory
Fil
e
in
per
spec
ti
v
e
of
the
thre
shold
val
ue
and
al
lo
cate
s
sec
ondar
y
m
emor
y
loc
at
ion
base
d
on
the
req
uire
m
ent.
Ac
cor
ding
to
th
e
pr
oposed
al
gor
it
h
m
the
m
axi
m
um
thi
rd
,
out
of
fourth
of
m
ai
n
m
emory
is
used
at
th
e
se
conda
r
y
f
il
e
cachi
ng
tim
e.
The
f
ree
spac
e
a
ids
in
fas
te
r
oper
at
ion
b
y
D
y
namic
al
l
y
Sc
al
ab
le
Nam
eNode
appr
oa
ch
.
Thi
s
proposed
a
l
gorit
hm
show
s
tha
t
th
e
spa
ce
u
tilizati
on
is
inc
r
eas
ed
to
17%
and
ti
m
e
utiliza
ti
on
is
inc
r
ea
se
d
to
0.
0005%
with
the
compa
rison
of
the
exi
sting
fra
gm
en
ta
ti
on
a
lgori
thm.
Ke
yw
or
ds:
DataN
od
e
Hado
op
Me
ta
data
Nam
eNo
de
Seco
nd
a
ry St
orage
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
:
Tum
pa
Ra
ni S
hah
a
,
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce a
nd E
ng
i
ne
erin
g,
Dh
a
ka U
niv
e
rs
it
y of
E
ng
i
neeri
ng
a
nd Tec
hn
ology,
Gazip
ur, Ban
glades
h.
Em
a
il
: t
u
m
pa.
cse4
9@gm
ail.co
m
1.
INTROD
U
CTION
In
t
his
m
od
er
n
age,
it
has
bec
om
e
the
m
ai
n
con
ce
r
n
to
ha
ndle
the
data
th
at
is
bein
g
generate
d
e
ver
y
day.
Appro
xim
at
el
y 25
qu
i
ntil
li
on
byte
s of dat
a are cr
eat
e
d every
day an
d 90% o
f
t
he
dat
a h
as
bee
n
cr
ea
te
d
in
the
la
st
two
ye
ars.
This
da
ta
are
bein
g
gen
e
ra
te
d
fro
m
ever
ywhe
re
li
ke
sensors
for
gat
her
in
g
cl
i
m
ate
inf
or
m
at
ion
,
soc
ia
l
m
edia
sit
e
s,
tran
sact
ion
r
ecords,
sat
el
li
t
es
et
c.
These
da
ta
set
s
are
i
m
m
ensely
un
str
uc
ture
d
and
as
a
resu
l
t
to
proces
s
a
nd
est
i
m
at
e
these
big
data
is
a
great
co
nce
rn.
As
the
dat
a
s
iz
e
has
i
ncrea
sed
extrem
el
y
RD
BM
S
has
f
ou
nd
it
c
halle
ngin
g.
M
or
e
e
ver
as
t
hese
data
set
s
are
sem
i
-
structu
r
ed
an
d
un
st
ru
ct
ur
e
d
RDBM
S
can
no
t cat
egorize
as
t
hey
are
desig
ne
d
to
ha
nd
le
struct
ur
e
d
data. Th
is
pro
blem
req
ui
res
a
database
m
a
nag
em
ent
syst
e
m
that
is
capab
le
of
a
naly
ze
these
data
in
an
e
ff
ic
ie
nt
and
c
onve
nien
t
way.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
1
3
, N
o.
2
,
Fe
bru
ary
201
9
:
7
2
9
–
7
3
6
730
Ap
ac
he
Ha
doop
is
su
ch
kind
of
DBMS
for
handlin
g
sem
i
-
structu
re
d
an
d
un
st
ru
ct
ur
e
d
da
ta
that
pr
ovid
es
a
n
op
e
n
s
ource
,
di
stribu
te
d
data
base
processi
ng
platf
orm
acro
ss
se
ver
al
t
housa
nd
nodes.
M
or
e
e
ver
it
is
high
sp
ee
d
a
nd
has
great
er
t
olera
nce
t
o
fau
lt
al
ong
with
cost
eff
ic
ie
ncy
as
i
t
stock
s
data
i
n
sm
al
l
a
m
ou
nt
via
m
ul
ti
ple serv
e
r
s.
Fil
e
syst
e
m
meta
data
and
a
ppli
cat
ion
data
are
store
d
se
pa
ratel
y
in
the
existi
ng
Ha
doop
Distri
bu
te
d
Fil
e
Syst
e
m
.
Nam
eNo
de
co
ntains
t
he
m
etad
at
a
of
t
he
s
yst
e
m
and
Dat
aNode
co
ntain
ap
plica
ti
on
da
ta
.
Pe
r
cl
us
te
r
ab
out
te
ns
of
th
ou
sa
nds
of
cl
ie
nts
can
acce
ss
the
H
adoo
p
sto
rag
e
at
a
tim
e.
DataNode
sto
re
the
blo
c
k
of
data
i
n
thei
r
local
file
syst
e
m
and
Nam
eNo
de
sto
re
m
et
a
d
at
a
of
al
l
the
DataN
od
e
in
t
heir
l
ocal
file
s
yst
e
m
.
So
i
f
we
try
t
o
exte
nd
the
ne
twork
or
try
t
o
ad
d
ne
w
DataNode
t
he
n
bec
ause
of
Nam
e
Node
m
ai
n
m
e
m
or
y
lim
it
at
ion
we
can’
t
e
xten
d
t
he
netw
ork.
T
his
na
m
espace
lim
i
ta
ti
on
is
on
e
of
t
he
im
portant
pro
blem
s
of
exi
sti
ng H
a
doop
Distrib
uted F
il
e Syst
e
m
.
This
pro
po
se
d
Dynam
ic
al
l
y
Scal
able
N
a
m
eNo
de
(
D
SN
)
ap
proac
h
intr
oducin
g
a
Me
ta
daya
Fr
a
gm
entat
ion
Algorithm
(M
FA
)
t
o
fr
a
gm
e
nt
the
m
e
ta
data
fr
eq
ue
ntly
and
inc
rease
th
e
nam
espace
capaci
ty
dynam
ic
al
l
y by
m
aking
t
he
i
nteract
io
n between
m
ai
n
m
e
m
or
y and
sec
onda
ry m
e
m
or
y of Nam
eNode.
2.
LIT
ERATUR
E REVIE
W
In
the
fiel
d
of
m
od
ern
te
ch
nolog
y,
the
us
e
of
Ha
doop
f
or
handlin
g
bi
g
da
ta
has
bec
ome
an
act
ive
area
of
resea
rc
h.
Seve
ral a
ppr
oach
e
s
hav
e
b
e
en
s
uggeste
d o
n
this
f
ie
ld.
In
[1
]
,
they
d
e
velo
ped
t
he
Ha
doop D
ist
ri
bu
t
ed
Fil
e Syst
e
m
o
n beh
al
f
of Ya
hoo.
T
hey h
a
ve
ex
plain
e
d
about the
H
a
do
op arc
hitec
ture
and s
howe
d
t
he
r
es
ult f
or
ha
ndli
ng 25 Pet
ab
yt
e o
f data
at
Yaho
o.
A
cl
assifi
cat
io
n
base
d
m
et
adata
m
anag
em
e
nt
syst
e
m
is
pr
opos
e
d
in
[
2].
They
f
ocu
se
d
on
reducin
g
the
bo
tt
le
nec
k
of
the
Nam
e
Node
m
ai
n
me
m
or
y.
They
f
rag
m
ent
the
m
et
adata
of
Na
m
eNo
de
based
on
the
i
m
po
rtance f
ac
tor.
T
hey
ha
ve
cal
culat
ed
thre
e
(H
ig
h,
Me
di
um
and
Lo
w)
t
ypes
of
im
po
rtance
fact
or
(
I
f).
Has
h
ta
ble
is
us
e
d
t
o
re
prese
nt
h
ig
h
I
f,
a
tree
m
ap
is
us
e
d
to
re
pr
ese
nt
m
edium
If
and
seq
ue
nce
file
s
are
us
e
d
t
o
represe
nt lo
w
I
f.
A
cache
d
a
ppr
oach
is
pro
po
s
ed
in
[
3]
f
or
a
ddressi
ng
Nam
eNode
scal
a
bili
ty
in
HD
FS
.
Their
m
ai
n
fo
c
us
was
to
e
nh
a
nce
the
e
xi
sti
ng
arc
hitec
ture
.
They
f
ra
gm
ent
t
he
m
et
a
data
de
pends
on
the
la
st
acce
ss
tim
e
and
m
ov
e
d
the
le
ast
fr
eq
ue
ntly
us
ed
data
to cache.
T
hey
w
ere
able
to r
em
ov
e 250
MB
o
f
data
from
RAM.
Bu
t
for
da
ta
searc
hi
ng
when
t
he
r
equ
e
ste
d
data
no
t
fou
nd
in
m
ai
n
m
e
m
or
y
then
the
seco
nda
ry
data
will
be
loade
d
a
gain
on
the
R
AM.
S
o
w
hen
the
sec
ondar
y
data
rel
oad
to
t
he
pr
im
ary
m
e
m
or
y
the
issue
of
Nam
eNode
m
a
in
m
e
m
or
y l
i
m
it
a
ti
on
a
rises a
gain.
In
pa
per
[
4],
they
analy
ze
t
he
requirem
ent
li
ke
ha
rdwa
r
e,
s
of
twa
re,
ne
twork
en
vir
onm
ent
fo
r
i
m
pr
ovin
g
the p
er
form
ance
of
cl
oud
c
om
pu
ti
ng
. Th
ey
d
eve
lop
e
d
a
cach
e
s
yst
e
m
in
la
ye
re
d
pas
sio
n
w
he
r
e
the
syst
e
m
has
a
cl
ie
nt
li
br
ary
a
nd
m
ulti
ple
cache
se
rv
ic
es.
Cl
ie
nt
li
br
ary
ca
n
acce
ss
t
he
fi
le
s
from
the
sh
are
d
m
e
m
or
y.
This
distrib
uted
ca
che
syst
em
ca
n
m
anipu
la
te
la
rg
e
num
ber
of
file
s
with
a
m
illi
secon
d
le
vel
in
highly
concu
rrent en
vir
onm
e
nt.
I
n
[
5],
they
de
velo
ped
a
m
ec
han
ism
to
i
m
pr
ov
e
d
Hado
op
perform
ance
usi
ng
m
et
adata
for
ha
ndli
ng
big
data.
By
as
sign
i
ng
jo
bs
to
the
DataN
od
e
,
H
2H
a
doop
w
as
exte
nd
e
d
t
he
abili
ty
of
Na
m
eNo
de
.
T
hey
wer
e
su
ccess
fu
l
f
or
r
edu
ci
ng CP
U
t
i
m
e and
nu
m
be
r of
nee
d op
e
r
at
ion
.
I
n
[
6
-
8],
they
pro
po
se
d
a
s
yst
e
m
fo
r
im
p
rovin
g
m
et
adata
m
anag
em
ent
in
HDFS
for
sm
a
ll
file
s.
They f
oc
us
e
d on the
sm
all f
il
es in t
he
m
ai
n
m
e
m
or
y and p
r
ov
i
de
a
rch
i
val
m
et
ho
ds f
or th
os
e sm
al
l fi
le
s.
Distrib
uted
m
et
adata
m
anag
e
m
ent
schem
e
is
propose
d
in
[
9].
They
pro
po
se
d
a
syst
e
m
fo
r
distrib
uted
m
et
adata m
anag
e
m
ent sch
em
e i
n HDFS t
o
im
pr
ove t
he HD
FS
eff
ic
ie
nc
y.
In
[10],
t
he
na
m
espace
is
de
par
te
d
into
sev
eral
fr
a
gm
ents.
Re
plica
s
of
ea
ch
f
ra
gm
ent
are
disp
e
rs
e
d
a
m
on
g
the
NN.
More
ti
m
e
is
need
e
d
f
or
m
et
adata
searchi
ng
with
sync
hro
nizat
ion
bec
ause
the
f
rag
m
ented
nam
espaces ar
e d
ist
rib
ute
d
a
m
on
g
di
ff
e
ren
t
NN
In
[11],
they
pro
po
se
d
a
D
ynam
ic
Director
y
Pa
rtit
ion
in
g
(
D
DP
)
te
c
hniq
ue
w
he
re
they
al
lowi
ng
directo
ry
m
e
tad
at
a
an
d
file
m
et
adata
in
a
div
e
rse
way.
They
i
m
pr
ove
d
the
pe
rfo
rm
a
nce
on
scal
abi
li
ty
and
adap
ta
bili
ty
.
An
ef
fici
ent
m
et
adata
m
anag
em
ent
syst
e
m
is
propose
d
in
.
T
hey
pro
po
s
ed
direct
ory
le
vel
base
d
m
et
adata
m
anag
em
ent
wh
ic
h
is
m
or
e
eff
ic
ie
nt
th
an
t
he
dire
ct
or
y
sub
tree
par
ti
ti
on
i
ng
a
nd
tradit
io
nal
ha
sh
in
g
te
chn
iq
ue.
3.
RESEA
R
CH MET
HO
D
The
D
SN
m
eth
od
ology
has
the
fo
ll
ow
i
ng
desig
n
pri
nci
pl
e
(1
)
Dynam
i
cal
ly
Scal
able
Nam
eNo
de
arch
it
ect
ure
an
d
(
2)
w
orki
ng
proce
dure.
In
this
sect
io
n,
t
he
syst
e
m
arch
it
ect
ur
e
a
nd
the
work
i
ng
pr
oce
dure
of
the DSN
arc
hitec
ture
is
give
n.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A no
ble ap
pr
oach
to devel
op
dyna
mically
sc
alable n
am
en
ode in
ha
doop
di
stri
bu
te
d…
(
T
ump
a
R
ani S
ha
ha
)
731
3.1
.
Dyna
mi
c
ally S
c
alabl
e
Na
m
eN
od
e
Ar
chitecture
The
Dynam
ic
a
ll
y
scal
able
N
a
m
eNo
de
arc
hi
te
ct
ur
e
is
s
hown
in
F
ig
ur
e
1.
D
NS
has
m
ast
er/sl
ave
arch
it
ect
ure.
DNS
cl
us
te
r
c
on
sist
s
of
a
sing
le
Na
m
eNode,
a
m
as
te
r
serv
e
r
that
m
a
nag
e
s
the
file
syst
e
m
nam
espace
and
regulat
es
acce
ss
to
file
s
by
cl
ie
nts
and
a
num
ber
of
Data
Node,
us
ually
on
e
per
no
de
in
the
cl
us
te
r.
I
n
ov
e
r
al
l,
the
DSN
s
yst
e
m
con
sist
s
of
on
e
Nam
e
Node,
a
gro
up
of
DataN
od
e
,
cl
ie
nts,
m
a
in
m
e
m
or
y
and sec
ondar
y
m
e
m
or
y con
ce
pt which
is
discuss
e
d
in
this s
ect
ion
.
Figure
1. Dy
na
m
ic
al
l
y Scal
able Nam
eNo
de Arc
hitec
ture
3.2
.
Na
m
eN
ode
The
f
ocal point
o
f HD
FS
is N
a
m
eNo
de
. I
t k
eeps th
e trac
k wh
e
re th
e
file
d
at
a is kep
t
over th
e cluste
r
.
The
directo
ry t
ree of all
f
il
e i
n
the
syst
em
al
so
kep
t
he
re. Wh
e
n
t
he
cl
ie
nt
s w
ish t
o
lo
cat
e a f
il
e or t
hey n
eed
to
add
/c
op
y/
delet
e/m
ov
e
a
file
t
hen
cl
ie
nt
ap
plica
ti
on
s
sen
d
a
req
ue
st
to
the
Nam
eNo
de.
T
he
Nam
eNo
de
rep
li
es
with c
orres
pondin
g Data
N
od
e
addres
s.
3.
3
.
M
ain
Me
mor
y
Norm
al
l
y
the
nam
espace
of
the
Ha
doop
syst
e
m
is
store
d
i
n
Nam
eNo
de
m
ai
n
m
e
m
or
y.
In
this
pro
po
se
d
a
rc
hi
te
ct
ur
e
we
in
tro
du
ce
Ma
in
Mem
or
y
Fil
e
(MM
F)
c
once
pt,
wh
ic
h
st
ores
the
high
pr
iority
m
et
adata of
t
he
syst
e
m
.
3.
4
.
Sec
on
d
ar
y
Mem
ory
In
t
his
propos
ed
a
rch
it
ect
ure
we
i
ntrod
uce
d
the
seco
ndar
y
m
e
m
or
y
concept.
T
he
fr
a
gm
ented
lo
w
pr
i
or
it
y
m
et
adata
will
store
in
the
sec
onda
ry
m
e
m
or
y.
A
lot
of
file
s
can
sto
re
in
th
e
seco
nd
a
ry
m
e
m
or
y
accor
ding t
o
th
e pro
posed
alg
or
it
hm
w
hic
h
i
s d
isc
us
se
d
in
wo
rk
i
ng
proce
dure s
ect
io
n.
3.
5
.
D
ata
Nod
e
DataN
od
e
cac
he
the
data
in
the
H
DF
S.
DataNode
ta
lks
to
the
Na
m
eNo
de
to
per
f
or
m
m
od
ific
at
ion
s
of
the
data
com
m
and
ed
by
the
Nam
eNo
de
an
d
res
ponse
to
the
Na
m
eNo
de
a
fter
a
fixed
ti
m
e
interval
con
ti
nu
ously
with
a
li
st
of
a
ch
unk
that
they
are
sto
r
ing
for
file
s
yst
e
m
activity.
Cl
ie
nts
syst
em
can
com
m
un
ic
at
e t
o
the
D
at
a
Nod
e d
irect
ly
if t
he
N
am
eNo
de
h
a
s assig
ne
d
the
address
of the
DataN
od
e
.
3.
6
.
Clie
n
ts
Cl
ie
nts
of
the
pro
po
se
d
syst
e
m
can
request
t
o
the
Nam
eNode
f
or
a
ny
par
t
ic
ular
file
.
Na
m
eNo
de
will
rep
ly
with
t
he
address
of
the
requested
Data
Node
to
the
cl
i
ents.
T
he
n
cl
ie
nts
dir
ect
ly
com
m
un
ic
at
e
with
the
DataN
od
e
for r
eadin
g or w
riti
ng ope
rati
on.
3.
7.
Me
t
adata
HDFS
m
et
adata
is
div
i
ded
in
to
tw
o
cat
eg
ori
es
of
file
s
na
m
ed
fsim
age
and
edits
lo
g.
T
he
c
om
plete
sta
te
of
the
file
processi
ng
s
yst
e
m
at
a
po
int
in
tim
e
is
c
on
te
nt
by
the
fsim
age
file
.
A
uniq
ue
inc
r
easi
ng
transacti
on
id
i
s
assig
ned
i
n
e
ver
y
m
od
ific
at
ion
of
file
syst
e
m
.
Af
te
r
al
l
m
od
ific
at
ion
to
that
id
fsim
a
ge
fil
es
represe
nts the f
il
e syst
e
m
stat
e
.
C
l
i
e
n
t
s
R
e
qu
e
s
t
R
e
ply
...
MMF
S
M
F
1
M
a
i
n
M
e
mo
r
y
S
e
c
o
n
d
a
r
y
M
e
mo
r
y
S
M
F
2
S
M
F
n
Da
ta
Node
…
Da
ta
Node
Da
ta
Node
Na
m
e
N
ode
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
1
3
, N
o.
2
,
Fe
bru
ary
201
9
:
7
2
9
–
7
3
6
732
3.
8.
W
orkin
g
Proced
ure
In
this
DSN
a
rch
it
ect
ure
w
he
n
Ha
doop
cl
ie
nt
re
qu
est
to
Nam
eNo
de,
it
firstly
check
t
he
avail
able
sp
ace
of
MM
F
(Mai
n
Me
m
or
y
Me
ta
data
Fil
e).
If
so
th
en
t
he
new
fil
e
is
create
d.
And
de
fau
lt
pri
or
it
y
1
(Lowest
P
rio
rity
)
is
set
fo
r
t
he
ne
wly
crea
te
d
file
.
But
if
there
is
no
a
vaila
ble
sp
ace
in
MM
F
then
le
ast
pr
i
or
it
y
m
et
adata
will
be
m
ov
e
d
to
SMF
(S
eco
ndary
Mem
or
y
Me
ta
data
Fil
e)
f
ollow
i
ng
t
he
pro
po
s
ed
Me
ta
data
Fr
ag
m
entat
ion
Algorithm
(MFA)
.
That
is
a
pr
ior
it
y
based
dyna
m
ic
m
et
adata
cl
assifi
er
is
propos
e
d
for
the
m
ai
n
m
e
m
or
y uti
li
zat
i
on. For assi
gn
i
ng prio
rity
let
u
s ass
um
e the foll
ow
i
ng p
a
ra
m
et
ers
T
d
=Fixe
d
Tim
e I
nterv
al
H=
Nu
m
ber
of H
it
s
durin
g
T
d
MM
S=M
ai
n
Mem
or
y Si
ze
M
th
=M
ai
n
Me
m
or
y Thr
es
ho
l
d
S
th
=Seco
nd
a
ry
Mem
or
y Th
res
ho
l
d
MM
F=M
ai
n
Mem
or
y M
et
adata Fil
e
SMF=Sec
on
da
ry Mem
or
y M
et
adata Fil
e
S= Size
of eac
h
m
et
adata
x= Num
ber
of
m
et
adata fil
e f
or M
th
= (MM
S/2)
/S
y=
N
um
ber
of
m
et
adata fil
e f
or S
th
= (MMS/
4)
/S
Gen
e
rall
y,
the
fu
ll
fsim
age
file
is
store
d
in
the
m
ai
n
m
e
m
or
y
of
Na
m
eNo
de
.
T
o
f
rag
m
ent
the
fsim
age
file
threshold
value
(Mth)
is
cal
cul
at
ed
by
(MM
S
)/2.
T
hat
is
ha
lf
of
t
he
m
ai
n
m
e
m
or
y
siz
e
i
s
th
e
thres
ho
l
d
f
or
MM
F.
Seco
ndary
Mem
or
y
Thr
es
hold
(
Sth
)
value
is
cal
cul
at
ed
by
(MM
S
)
/4.
S
o
x
is
the
nu
m
be
r
of
m
et
adata
fil
e
that
can
be
store
d
on
Mt
h
and
y
is
the
num
ber
of
m
et
a
data
file
that
c
an
be
sto
red
on
Sth.
Figure
2
s
hows
the m
et
adata fr
agm
entat
ion
a
lgorit
hm
.
Figure
2. Me
ta
data F
rag
m
entat
ion
Algorith
m
Wh
e
n
the
siz
e
of
the
m
e
ta
da
ta
file
exceeds
the
Mt
h
then
the
fr
a
gm
entation
al
gorithm
i
s
trigg
e
red.
Wh
e
n
t
he
th
re
sh
ol
d
value
e
xc
eeds,
t
he
n
th
e
pr
i
or
it
y
val
ue
for
eac
h
m
etad
at
a
will
be
updated
f
reque
ntly
if
need
e
d
ba
sed
on
trig
ge
r.
Ne
wly
gen
er
at
ed
pr
io
rity
values
are
so
rte
d
(hi
gh
e
r
to
lowe
r
or
de
r)
a
nd
m
et
adata
hav
i
ng h
i
gh
e
r pr
i
or
it
y wil
l ke
ep
to
the MMF
. T
hat is x n
umber
of m
et
adata has
b
ee
n
st
ored in M
MF
Lo
w
pri
ori
ty
m
et
adata
reco
r
ds
are
sepa
rat
ed
out
an
d
m
ov
e
d
int
o
the
file
create
d
on
sec
ondar
y
stora
ge.
A
s
lo
w
pri
or
it
y
m
etad
at
a
fr
e
quentl
y
m
ov
es
to
the
secondary
st
or
a
ge
so
the
num
ber
of
SM
F
will
exten
d
acc
ordi
ng
to
t
he
siz
e
of
m
et
adata.
T
he
nu
m
ber
of
m
et
adata
has
be
en
st
or
e
d
in
e
ach
SMF
is
m
easur
e
d
by
facto
r
y
and
they
m
us
t
be
store
d
acco
rd
i
ng
their
higher
t
o
lowe
r
pri
ori
ty
.
Let
con
side
r
the
siz
e
of
the
m
ai
n
m
e
m
or
y
is
1
GB,
then
the
t
hresh
old
val
ue
(
Mt
h)
will
be
512
MB
a
nd
the
siz
e
of
eac
h
fragm
ented
file
in
t
he
seco
nd
a
ry
m
e
m
or
y
(S
th)
is
1
GB/4=2
56
MB
.
If
we
c
on
sider
t
hat
siz
e
of
eac
h
m
et
adata
is
1MB
the
n
MM
F
can c
on
ta
i
n 512 m
et
adata
wh
i
ch
is
facto
r x.
Wh
e
n
t
he
us
er
search
e
s
a
ny
pa
rtic
ular
file
,
the
syst
em
will
search
that dat
a
in
the
m
ai
n
m
e
m
or
y
first.
If
it
is
found,
t
he
file
will
be
rep
li
ed
t
o
the
use
r
with
t
he
D
at
aNode
ad
dre
ss.
But
if
it
is
no
t
f
ound
in
t
he
m
ai
n
1
.
I
f
MM
F >
M
t
h
th
e
n
2.
C
alcu
late
n
e
w
P
r
io
r
ity
v
al
u
e
(
P
)
=
A
v
er
ag
e
(
Old
P
r
i
o
r
ity
,
H)
3.
So
r
t th
e
m
etad
ata
d
ep
en
d
in
g
o
n
P
in
d
escen
d
i
n
g
o
r
d
er
4.
Kee
p
h
ig
h
o
r
d
er
x
f
ac
to
r
o
f
d
ata
in
MM
F
5.
Sh
if
t r
es
t lo
w
e
s
t d
ata
to
SM
F [
i=1
….
n
]
6.
I
f
SMF[
i]
>
S
t
h
th
e
n
7.
R
ep
ea
t step
2
&
3
8.
Kee
p
h
ig
h
o
r
d
er
y
f
ac
to
r
o
f
d
ata
in
SMF[
i]
9.
Sh
if
t r
es
t lo
w
f
ac
to
r
d
ata
to
SMF[
i+1
]
10.
en
d
if
1
1
.
en
d
i
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A no
ble ap
pr
oach
to devel
op
dyna
mically
sc
alable n
am
en
ode in
ha
doop
di
stri
bu
te
d…
(
T
ump
a
R
ani S
ha
ha
)
733
m
e
m
or
y
then
a
ccord
in
g
to
t
he
pr
io
rity
value
the
re
qu
est
e
d
f
il
e
will
be
cached
t
o
the
m
ain
m
e
m
or
y
fr
om
the
seco
nd
a
ry m
e
m
or
y t
hr
ough
pag
e
table
wh
i
ch
is s
how
n
i
n Fi
gure
3.
Figure
3. Sec
onda
ry Fil
e Cac
hing
4.
RESU
LT
S
AND A
N
ALYSIS
To
e
valuate
the
perform
ance
of
the
MF
A
al
go
rithm
we
hav
e
c
onduct
ed
t
w
o
kinds
of
te
st:
1.
Perfo
rm
ance
on
m
ai
n
m
e
m
o
ry
us
a
ges
2.
P
erfor
m
ance
on
ave
rag
e
res
po
ns
e
ti
m
e.
In
t
his
sect
io
n
we
ha
ve
dem
on
strat
ed
the
perform
ance of th
e
DSN
a
ppr
oach an
d
t
he
co
m
par
iso
n wit
h
the
ex
ist
i
ng cache
app
roach.
4.1
.
Simul
at
io
n Pla
tform
We
ha
ve
de
ve
lop
e
d
the
MF
A
an
d
e
xisti
ng
f
rag
m
entat
ion
al
gorithm
usi
ng
C+
+
la
ng
uag
e
i
n
tw
o
diff
e
re
nt
com
pu
te
rs.
O
ne
of
t
ho
s
e
is
4G
B
RAM
wit
h
2.10
G
Hz
C
or
e
i
3
process
or
an
d
ano
t
her
one
is
8G
B
RAM wit
h 1
.60GHz C
or
e
i5 pr
ocess
or.
4
.
2
.
Per
fo
r
m
an
ce
on
M
ain
Mem
ory U
s
ag
es
In
t
his
sect
io
n
the
perform
a
nces
on
m
ai
n
m
e
m
or
y
us
age
s
of
DSN
a
ppro
ac
h
a
nd
e
xisti
ng
cac
he
d
appr
oach
i
n
te
r
m
s
of
siz
e
of
m
ai
n
m
e
m
or
y
is
discusse
d.
F
igure
4
sho
ws
the
Nam
eNode
m
a
in
m
e
m
or
y
us
a
ge
com
par
ison.
0
512
1024
1536
2048
2560
3072
3584
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
3000
3200
3400
3600
Us
e
d
of RAM (MB)
Size
of
RAM(MB)
Dy
nam
ically
Scalable
Nam
eNode
Approach
Ex
isting
Cached
Approach
Figure
4. Nam
eNode Mai
n
M
e
m
or
y Usa
ge C
om
par
ison
MMF
SMF1
SMF1
SMF2
SMFn
Ma
in
Me
m
o
r
y
A
d
d
r
ess
T
r
an
s
latio
n
1
C
ac
h
ed
0
0
Me
m
o
r
y
R
e
s
id
en
t
P
ag
e
T
a
b
le
Seco
n
d
ar
y
Me
m
o
r
y
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
1
3
, N
o.
2
,
Fe
bru
ary
201
9
:
7
2
9
–
7
3
6
734
Accor
ding
to
the
MFA
t
he
us
e
of
R
AM
for
the
Dyna
m
ic
al
l
y
Scal
able
Nam
eNo
de
appr
oach
is
cal
culat
ed
by
the
siz
e
of
x
fa
ct
or
,
y
fact
or
a
nd
the
siz
e
of
each
m
et
adata.
Af
te
r
the
f
ra
gm
entat
ion
of
c
ach
e
appr
oach
t
he
m
ai
n
m
e
m
or
y
can
sto
re
700M
B
m
et
adata
and
25
0MB
da
ta
in
the
sec
on
dar
y
m
e
m
or
y
of
1GB
RAM
[3
]
.
But
in
DS
N
syst
e
m
the
m
a
in
m
e
m
or
y
is
able
to
ho
l
d
512M
B
and
25
6MB
in
secondary
m
e
m
or
y
after
the
m
et
a
data
fr
a
gm
entat
ion
al
gorithm
trigg
e
r.
T
he
Seco
nd
a
ry
Me
m
or
y
can
store
sever
al
file
s
of
siz
e
256MB.
So t
he
stor
a
ge
ca
paci
ty
h
as
been inc
reased
d
y
nam
i
cal
ly
.
Existi
ng
cac
he
d
ap
proac
h
is
us
e
d
92%
of
R
AM
an
d
the
D
SN
al
gorithm
req
ui
red
m
axi
m
um
75
%
of
m
ai
n
m
e
m
or
y
in
w
orst
case.
So
t
his
D
SN
a
ppr
oach
is
util
iz
ed
a
v
era
ge
17%
of
m
ai
n
m
e
m
or
y
us
age.
T
his
f
ree
sp
ace
of m
ai
n
m
e
m
or
y ensur
e the
ov
e
rall
r
e
sp
onse
tim
e o
f t
he
Nam
eNode
.
4
.
3
.
Per
fo
r
m
an
ce
on
Re
sponse Ti
me
In
this
sect
io
n
the
perf
or
m
ances
on
a
ver
a
ge
res
pons
e
ti
m
e
of
DSN
a
ppr
oac
h
an
d
ex
ist
ing
cache
d
appr
oach
is
dis
cusse
d.
F
or
a
na
ly
zi
ng
the
av
erag
e
re
spo
ns
e
tim
e
of
the
N
a
m
eNo
de
,
we
hav
e
m
ade
a
set
up
to
si
m
ulate
of
pr
opos
e
d
a
nd
e
xi
sti
ng
MF
A
al
gorithm
in
tw
o
well
co
nfi
gu
red
com
pu
te
rs
.
Setu
p
-
1:
4GB
RAM
with
2.1
0
G
Hz
Core
i3
proce
sso
r
a
nd
Set
up
-
2:
8G
B
RA
M
with
1.
60
G
Hz
Core
i5
process
or.
Figur
e
5
an
d
Figure
6
s
how
the av
e
ra
ge res
pons
e
tim
e analy
sis of
set
up
-
1 an
d
set
up
-
2.
0
512
1024
1536
2048
2560
3072
3584
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.10
0.11
0.12
R
esp
o
n
se Ti
me in
Secon
d
s
(
S)
S
ize
of
R
AM
(
MB
)
Dy
na
mi
c
a
ll
y
S
c
a
la
ble
Nam
e
Node A
ppr
oa
c
h
Ex
is
ti
ng
C
a
c
he
Appr
oa
c
h
Figure
5.
A
verage Res
pons
e
Ti
m
e A
naly
sis
of Setu
p
-
1
Let
con
si
der
th
e
siz
e
of
each
m
et
adata
(S
)
is
1MB.
The
n
th
e
MM
F
will
con
ta
in
51
2
m
e
ta
data
w
hich
is facto
r X a
nd each SM
F ca
n co
ntain
256
m
et
adata w
hich
is facto
r Y.
0
512
1024
1536
2048
2560
3072
3584
0
.
0
0
0
.
0
1
0
.
0
2
0
.
0
3
0
.
0
4
0
.
0
5
0
.
0
6
0
.
0
7
0
.
0
8
0
.
0
9
0
.
1
0
0
.
1
1
0
.
1
2
Re
sp
o
n
se
Tim
e
in
Se
c
o
n
d
s
(S
)
Size
of
RAM(MB)
Dy
nam
ically
Scalable
Nam
eNode
Approach
Ex
isting
Cached
Approach
Figure
6. A
verage Res
pons
e
Ti
m
e A
naly
sis
of setu
p
-
2
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A no
ble ap
pr
oach
to devel
op
dyna
mically
sc
alable n
am
en
ode in
ha
doop
di
stri
bu
te
d…
(
T
ump
a
R
ani S
ha
ha
)
735
In
sim
ul
at
ion
,
the
config
ur
at
i
on
of
set
up
-
2
is
hig
he
r
than
set
up
-
1.
So
we
can
see
that
t
he
ave
rag
e
respo
ns
e
tim
e
in
set
up
-
2
is
le
ss
than
set
up
-
1.
So
her
e
it
is
pr
ove
d
that
this
propose
d
syst
e
m
wil
l
pr
ovide
bette
r
re
spo
ns
e
tim
e in h
ig
h
c
onfig
ur
e
d sy
ste
m
.
5.
CONCL
U
SI
O
N
In
t
his
propose
d
work
we
ha
ve
e
xp
e
rim
ent
ed
with
a
la
r
ge
am
ou
nt
of
da
ta
eff
ic
ie
ntly
thu
s
t
he
ti
m
e
requirem
ents
ha
s
bee
n
reduce
d
a
nd
m
e
m
or
y
util
iz
at
ion
is
i
ncr
ease
d.
T
he
pro
po
se
d
syst
em
is
m
or
e
ef
fi
ci
ent
than
the
e
xisti
ng
cache
d
a
ppr
oa
ch
that
is
pro
ved
by
ou
r
pe
r
form
ance
evaluati
on
secti
on. By
i
m
ple
m
entin
g
t
he
con
ce
pt
of
se
conda
ry
stora
ge
it
has
been
sh
ow
n
that
am
ou
nt
of
m
etad
at
a
will
no
t
be
so
hi
gh
t
hat
the
Nam
eNo
de
will
be
irres
pons
i
ve
due
to
the
e
xcessive
am
ount
of
data.
At
t
he
sam
e
t
i
m
e
t
he
cl
ie
nt
r
eq
ue
st
can
be
handled
m
or
e
fr
e
quently
than
t
he
e
xisti
ng
syst
em
.
In
fu
t
ur
e
w
ork
we
would
li
ke
to
intr
oduce
sever
al
par
am
et
ers
an
d
be
pro
ve
d
m
at
he
m
at
ic
a
lly
so
that
the
syst
e
m
can
work
m
or
e
ef
fici
ently
and
can
be
i
m
ple
m
ented
in
real t
i
m
e syste
m
.
REFERE
NCE
S
[1]
K.
Shvachko,
H.
Kuang,
S.
Radi
a,
and
R.
Chansl
er,
“
The
Hadoop
Distribut
ed
File
Sy
st
em
,
”
IEEE
26th
Symposium
,
pp.
1
–
10
,
Ma
y
,
2010
.
[2]
A.Cha
ndra
sek
ar,
K.Cha
ndra
seka
r,
H.
Ramasatagopan,
and
J.B
al
asubra
m
aniy
a
n
,
“
Cla
ss
ifi
ca
t
ion
base
d
Meta
da
t
a
Mana
gement
f
or
HD
F
S,”
IEE
E
14th
Inte
r
nati
onal
Confer
enc
e
on
High
Pe
rform
ance
Computing
and
Comm
unic
ati
ons,
2012
[3]
Zha
ng,
G.
W
u,
X.
Hu,
and
X.
W
u,
“
A
Distribut
ed
Cac
h
e
for
Hadoop
Distribut
e
d
File
Sy
st
em
in
Real
-
ti
m
e
Clou
d
Serv
ic
es,
”
13
th
I
nte
rnational
Co
nfe
renc
e
on
Gr
id
Computing, pp
.
12
-
21
,
2012
.
[4]
H.
Alsham
m
ari
,
J.L
ee,
and
H.
B
aj
wa,
“
H2H
adoop:
Im
proving
Hadoop
Perform
anc
e
using
the
Me
ta
da
ta
of
relate
d
jobs,”
I
EEE
Tr
ansacti
ons on
C
lo
ud
Computing
,
PP
.
1
-
1,
2015
.
[5]
G.
Mac
ke
y
,
S.
S
ehr
ish,
and
J
.
W
ang,
“
Im
proving
Meta
d
at
a
Mana
gement
f
or
Sm
al
l
File
s
in
HD
FS
,
”
IEEE
Inte
rnational
Co
nfe
renc
e
on
Clus
te
r Computing
a
nd
Workshops
,
2
009.
[6]
S.
Bende,
R.
Sh
edge
,
“
Dea
li
ng
with
Sm
al
l
Files
Problem
in
Hadoop
Distributed
File
S
y
stem,
”
7th
Int
ernati
on
a
l
Confe
renc
e
on
C
omm
unic
ati
on,
Computing
and
Vi
rtuali
zat
ion, 2
016.
[7]
Dr.
Raut
,
S
.
,
Phakade
,
P.
,
“
An
Innova
ti
ve
Strat
eg
y
for
Im
prove
d
Proce
ss
i
ng
of
Sm
al
l
File
s
in
Hadoop”
Inte
rnational
Jo
urnal
of Appl
i
ca
ti
on
or Inno
vat
io
n
in Engi
ne
ering
and
Manag
eme
nt
,
3
,
pp
.
278
-
28
0,
Jul
y
,
2014.
[8]
M.
Vara
de
,
V.Je
tha
ni
,
“
Distributed
Meta
d
at
a
Ma
nage
m
ent
Sche
m
e
in
HD
FS
,
”
Inte
rnational
Jou
rnal
of
Sci
en
ti
f
i
c
and
Re
search
P
ubli
cations
,
3
,
2
013.
[9]
Y.
KIM
,
T
.
Ar
ara
gi
,
J.
Naka
m
ura
,
T.
Masuza
wa,
“
A
Distribu
te
d
Nam
eNode
Cluste
r
for
a
H
ighly
-
Avai
la
b
le
Hadoop
Distribu
te
d
Fil
e
S
y
stem,
”
IE
EE 33rd Inte
rnational
S
ymposium on
Reliable
Distribute
d
Syst
em
,
2014
.
[10]
Y.
Fu,
N.
Xiao,
and
E.
Zhou,
“
A
Novel
Dy
na
m
ic
Meta
data
Mana
gement
Sc
heme
for
La
rge
Distribut
ed
Stor
age
S
y
stems
,
”
10
th
I
EE
E
Inte
rnat
ion
al
Conf
ere
nce o
n
High
P
erformance
Comput
ing
and
Comm
unications
,
2008
.
[11]
L.
R
an,
and
H
.
Jin,
“
An
Eff
icient
Me
ta
d
at
a
Mana
gement
M
et
hod
in
L
arg
e
Distribut
ed
St
ora
ge
S
y
s
te
m
s,
”
Inte
rnational
Co
nfe
renc
e
on
Hu
man
-
ce
ntric
Co
mputing
and
Embe
dded
and
Mul
ti
media
Comput
i
ng,
pp.
375
-
383,
20
11
.
BIOGR
AP
HI
ES OF
A
UTH
ORS
Tu
mpa
Ra
ni
S
hah
a
obta
ine
d
h
er
B.
Sc
and
M.
Sc
degr
ee
s
unde
r
depa
rtment
of
Com
pute
r
Scie
nc
e
and
En
gine
er
ing
from
Dhaka
Univer
sit
y
of
Engi
ne
eri
n
g
and
Technol
o
g
y
(DU
ET
)
,
Gaz
ipur,
Bang
ladesh.
Tumpa
Rani
shaha
rese
ar
c
h
int
ere
sts
are
o
n
Data
Mining,
Big
Data
,
Hadoop
Distribu
te
d
Fil
e
S
y
s
te
m
,
Mac
hin
e
L
ea
rn
i
ng
and
De
ep
Learni
ng.
Curre
n
tl
t
she
is
a
fac
ul
t
y
m
ember
at
d
epa
rtment
of
Com
pute
r
Sci
en
ce
and
Eng
ineeri
ng,
Daffodi
l
In
ternat
ion
al
Univer
sit
y
.
Md.
Nas
im
A
kh
tar
rec
ei
ved
the
M.E
ng
an
d
Ph.D
degr
ee
s
from
Nati
onal
Te
chnica
l
Univer
sit
y
of
Ukrai
ne,
Ki
ev,
Ukrai
ne
and
M
oscow
Stat
e
Aca
dem
y
o
f
Fine
Chemica
l
Te
chno
log
y
,
Ru
ss
ia
,
in
1998
an
d
2010,
respe
cti
vely
.
Cur
ren
t
l
y
,
he
is
a
Profess
or
in
th
e
Depa
rtment
of
Com
pute
r
Scie
nce
and
Engi
n
ee
r
ing,
Dhaka
Uni
ver
sit
y
of
Eng
in
ee
ring
and
Te
chno
log
y
(D
UET)
,
Gaz
ipur,
Bangl
ad
esh.
His
rese
arc
h
in
te
r
ests
inc
lude
Distri
bute
d
Dat
a
W
are
house
S
y
st
em
On
La
rg
e
Cl
usters,
Digi
ta
l
I
m
age
Proce
ss
in
g
and
W
ater
Ma
rking,
Pe
er
to
Peer
Networ
king,
C
loud
Co
m
puti
ng,
Opera
t
ing
S
y
stem.
He
has
pre
sent
ed
pape
rs
a
t
conf
ere
n
ce
s bo
th
hom
e
and
abr
o
a
d,
publ
ished art
i
cl
es
and
p
ape
rs
i
n
var
ious
journal
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
1
3
, N
o.
2
,
Fe
bru
ary
201
9
:
7
2
9
–
7
3
6
736
Md.
Z
akir
Ho
ss
ain
rec
ei
v
ed
t
he
B.
Sc
Engi
n
ee
ring
degr
ee
i
n
Com
pute
r
Sc
ie
nc
e
and
Engi
ne
eri
ng
De
par
tment
from
Dhaka
Univer
sit
y
of
Engi
ne
eri
n
g
and
Technol
o
g
y
(DU
ET
)
,
Gaz
ipur,
B
angl
a
desh,
in
2015
an
d
he
is
cur
ren
tly
pursuing
the
M.Sc
Engi
ne
eri
ng
degr
ee
in
Com
pute
r
Scie
n
ce
and
Engi
ne
er
ing
Depa
rtment
in
Dhaka
Univ
ersity
of
Eng
ineeri
ng
and
Te
chno
log
y
(DU
ET
),
Gaz
ipur.
His
rese
arc
h
in
t
ere
st
in
cl
udes
D
at
a
Mining
,
Big
Data
,
AI,
Mac
hine
Learni
ng,
Cloud
Com
puti
ng,
Softwar
e
Engi
ne
eri
ng
,
C
om
pute
r
Network,
IoT
.
He
has
pre
sent
ed
p
a
per
s a
t
conf
ere
n
ce
s bot
h
hom
e a
nd
abr
oad
Ms.
Fate
ma
Tu
j
Joohora
is
the
Le
c
ture
r
of
a
rep
ute
d
private
uni
ver
sit
y
in
Banglade
sh.
Sh
e
has
rec
ei
v
ed
her
B.
Sc
and
M.Sc
degr
ee
in
Infor
m
at
ion
Te
chn
olog
y
from
Jaha
ngirna
g
ar
U
nive
rsit
y
(JU
).
Her
rec
ent
publ
i
ca
t
ions
inc
lude
“
An
Eff
ic
ie
nt
Approac
h
of
Tra
in
ing
Artif
ici
al
Neura
l
Net
wo
rk
to
Rec
ognize
Benga
li
Hand
Sign"
(2016).
H
er
rese
ar
ch
int
er
est
inc
lud
e
s
Data
Mining,
Artifi
cial
Neu
r
al
Ne
twork,
Im
a
ge
proc
essing,
and
cl
oud
computing.
Mostafij
ur
Rah
man
complet
ed
his
BS
c
in
Comput
er
Sci
enc
e
fr
om
Nati
onal
Univer
sit
y
o
f
Bangl
ad
esh
(2003).
He
Purs
ued
his
MS
c
(2009)
and
PhD
(2017)
in
Com
pute
r
Engi
nee
r
ing,
from
UN
IM
AP
,
Malay
s
ia.
He
w
orke
d
as
L
ec
tur
e
r
since
2009
to
Septe
m
ber
,
2017
for
School
of
Com
pute
r
an
d
Com
m
unic
at
i
on
Engi
n
ee
ring
in
UN
IMA
P.
Curre
ntly
he
is
serving
as
As
sis
ta
nt
Profes
sor
in
the
Depa
rtment
of
Software
Engi
n
ee
r
ing
at
Daffodi
l
Int
ern
ational
Univer
sit
y
(DIU
),
Bangl
ad
esh.
His
rese
arc
h
intere
st
in
Softwar
e
Te
sting
,
Multi
m
edi
a
and
Crea
t
ivi
t
y
in
Medic
a
l
Scie
nc
e,
Com
pute
r
Secur
ity
,
C
lou
d
Com
puti
ng,
Algorit
hm
Optimiza
ti
o
n,
P
ara
l
le
l
and
Distribut
ed
S
y
stem
,
Device
Drive
r
for
GN
U/Li
nux
base
d
embedde
d
OS
.
He
has
pre
sented
pape
rs
at
conf
ere
nc
es
both
hom
e
and
abr
oad,
publi
shed
art
i
cl
es
and
p
apers
in
var
ious j
our
nal
s.
R.
Bad
li
sh
ah
A
hmad
rec
ei
ved
t
he
M.Sc
Engi
ne
eri
ng
and
Ph.D
degr
ee
s
from
Univer
sit
y
of
Strat
hcly
d
e,
UK
in
1995
and
20
00,
respe
ctively
.
Curre
nt
l
y
,
h
e
is
a
Fa
cul
t
y
of
In
form
at
ic
s
and
Com
puti
ng,
Univer
siti
Sult
an
Zaina
l
Abid
in
(UniSZA).
His
rese
arc
h
in
te
rests
in
Com
pute
r
and
T
el
e
comm
unic
at
i
on
Net
work
Modell
ing
include
W
SN
and
Optic
al
Networ
k
using
discre
t
e
e
vent
sim
ula
tors
(OM
NeT
++
),
O
pti
c
al
Networki
ng
and
Embedd
ed
S
y
stem
base
d
on
GN
U/Li
nux.
He
has
pre
sente
d
p
ape
r
s
at
conf
ere
nc
e
s
both
hom
e
a
nd
abr
oad
,
publi
shed
art
i
cle
s a
nd
pap
ers
in
v
ari
ous j
ourn
al
s.
Evaluation Warning : The document was created with Spire.PDF for Python.