Inter
national
J
our
nal
of
Recongurable
and
Embedded
Systems
(IJRES)
V
ol.
15,
No.
3,
No
v
ember
2026,
pp.
819
∼
836
ISSN:
2089-4864,
DOI:
10.11591/ijres.v15.i3.pp819-836
❒
819
Heter
ogeneous
inter
net
of
things
de
vice
inter
operability
using
f
og-based
adapti
v
e
semantic
mechanism
De
v
amekalai
Nagasundaram
1
,
Selv
akumar
Manickam
1
,
Mahmood
A.
Al-Shar
eeda
2,3
,
Huda
Mohammed
Alsay
ednoor
4
,
Ali
A.
Y
assin
5
,
Shankar
Karuppayah
1
1
Cybersecurity
Research
Centre
(CYRES),
Uni
v
ersiti
Sains
Malaysia,
Pulau
Pinang,
Malaysia
2
Department
of
Computer
Engineering
T
echnologies
and
Articial
Intelligence,
Basrah
Engineering
Polytechnic
Colle
ge,
Southern
T
echnical
Uni
v
ersity
,
Basra,
Iraq
3
Colle
ge
of
Engineering,
Al-A
yen
Uni
v
ersity
,
Thi-Qar
,
Iraq
4
Shatt
Al-Arab
Uni
v
ersity
Colle
ge,
Basra,
Iraq
5
Department
of
Computer
Science,
Colle
ge
of
Education
for
Pure
Sciences,
Uni
v
ersity
of
Basrah,
Basrah,
Iraq
Article
Inf
o
Article
history:
Recei
v
ed
Apr
21,
2026
Re
vised
Jun
23,
2026
Accepted
Sep
11,
2026
K
eyw
ords:
Adapti
v
e
semantic
mechanism
F
og
computing
Internet
of
things
Ontology
enrichment
Resource
description
frame
w
ork
Semantic
interoperability
ABSTRA
CT
W
ith
the
rapid
gro
wth
of
heterogeneous
internet
of
things
(IoT)
ecosystems,
insuf
cient
semantic
interoperability
caused
by
non-uniform
communication
pro-
tocols,
data
formats,
a
nd
semantic
representations
has
become
a
core
challenge
in
the
eld.
Existing
cloud-based,
static
ontology
schemes
ha
v
e
defects
including
high
latenc
y
,
lar
ge
o
v
erhead,
and
poor
adapt
ability
.
This
paper
proposes
the
dynamic
semantic
fog
(DSF)
fram
e
w
ork,
which
performs
semantic
processing
at
the
fog
layer
to
achie
v
e
real-time
interoperability
.
The
frame
w
ork
inte
grates
four
core
modules:
w
orkload
of
oading,
resource
description
frame
w
ork
(RDF)
generation,
semantic
representation,
and
adapti
v
e
ontology
e
xpansion.
It
sup-
ports
dynamic
access
to
ne
wly
added
IoT
de
vices.
T
ests
were
completed
using
iF
ogSim
in
a
smart
city
traf
c
scenario.
Compared
with
traditional
cloud
solu-
tions,
the
frame
w
ork
re
duces
latenc
y
,
ener
gy
consumption,
and
netw
ork
usage
rate
by
51.7%,
41.3%,
and
55.6%
respecti
v
ely
,
and
can
support
dynamic
semantic
interoperability
for
heterogeneous
IoT
en
vironments.
This
is
an
open
access
article
under
the
CC
BY
-SA
license
.
Corresponding
A
uthor:
Selv
akumar
Manickam
Cybersecurity
Research
Centre
(CYRES),
Uni
v
ersiti
Sains
Malaysia
Pulau
Pinang,
11800,
Malaysia
Email:
selv
a@usm.my
1.
INTR
ODUCTION
W
ith
the
rapid
de
v
elopments
of
internet
of
things
(IoT)
paradigms,
billions
of
heterogeneous
de
vices
ha
v
e
been
deplo
yed
in
all
kinds
of
application
elds,
including
smart
cities,
inte
lligent
transportation
syst
ems,
healthcare,
and
industria
l
automation
[
1
]–[
3
].
The
y
are
v
ery
dif
ferent
in
their
hardw
are
resources,
communication
protocols,
data
formats,
and
semantic.
While
this
di
v
ersity
allo
ws
for
e
xible
and
customized
IoT
deplo
yments,
it
also
poses
enormous
challenges
in
realizing
seamless
interoperability
between
de
vices
and
services
[
4
]–[
6
].
One
of
the
k
e
y
issues
that
hinder
scalability
and
decision
making
in
real-time
is
semantic
interoperability
,
which
refers
to
IoT
systems’
capabilit
y
to
understand
and
reuse
unambiguously
data
collected
from
di
v
erse
sources
[
7
],
[8].
Semantic
interoperability
attempts
to
guarantee
that
shared
data
has
syntactic
compatibility
as
well
as
semantics
for
all
in
v
olv
ed
entities
within
the
system
[
9
],
[
10
].
Ontology-based
semantic
models
utilizing
J
ournal
homepage:
http://ijr
es.iaescor
e
.com
Evaluation Warning : The document was created with Spire.PDF for Python.
820
❒
ISSN:
2089-4864
resource
description
frame
w
ork
(RDF)
and
web
ontology
language
(O
Wl),
pro
vide
rich
e
xpressi
v
eness
and
ha
v
e
been
lar
gely
used
to
combat
such
a
challenge.
Ne
v
ertheless,
the
majority
of
current
solutions
for
semantic
interoperability
tak
e
as
a
premise
static
en
vironments,
in
which
de
vice
types
are
pre-dened
and
the
ontologies
are
frozen
[
10
],
[
11
].
Assuming
as
such
is
not
practical
for
modern
IoT
ecosystems,
gi
v
en
that
there
is
a
dynamic
introduction
of
ne
w
de
vices,
data
attrib
utes,
and
services
occurring
[
12
].
In
addition,
most
of
t
he
e
xisting
semantic
interoperability
frame
w
orks
use
cloud-centric
architecture
where
centralised
cloud
se
rv
er
is
responsible
to
do
the
semantic
annotation
and
reasoning
[
13
],
[
14
].
While
the
cloud
platforms
of
fer
a
lot
of
a
v
ailable
computing
resources,
the
y
produce
service
latencies,
netw
ork
o
v
erheads
and
ener
gy
consumption
which
are
not
suitable
for
light-latenc
y-sensiti
v
e
and
real-time
IoT
deplo
yments
[15],
[16].
F
og
computing
has
recently
raised
as
a
ne
w
paradigm
to
reply
t
hese
chall
enges
by
shifting
cloud
po
wer
closer
to
the
data
sources
[
17
],
[
18
].
F
og
computing,
by
dis
trib
uting
computation
and
storage
capacities
to
the
netw
ork
edge,
allo
ws
for
lo
w-latenc
y
processing
and
minimizes
traf
c
to
the
cloud.
Ho
we
v
er
,
despite
these
merits,
the
state-of-the-art
fog-assisted
s
emantic
interoperability
solutions
still
lack
a
simple
and
ef
fecti
v
e
w
ay
to
manage
dynamic
adaptations
in
semantics
[
19
],
[
20
].
In
most
cases,
fog
nodes
to
only
do
partial
processing
and
ontology
handling
is
centralized
or
static,
which
leads
to
lack
of
scalable
and
adapti
v
e
in
heterogeneous
IoT
.
Although
much
research
progress
has
been
achie
v
ed
for
the
semantic
interoperability
of
heterogeneous
IoT
systems,
relying
on
tw
o
mainstream
approaches—ontology-based
and
fog-assisted
methods—core
unre-
solv
ed
limitations
still
persist:
static
ontologies
require
manual
adaptation
to
support
ne
wly
added
de
vices
;
cloud-centric
schemes
suf
fer
from
the
problems
of
high
latenc
y
,
high
o
v
erhead,
and
high
ener
gy
consumption;
and
the
ontology
e
v
olution
of
emer
ging
fog-assisted
approaches
still
has
k
e
y
a
ws,
remaining
either
static
or
o
v
erly
reliant
on
cloud-side
resources.
T
o
address
the
limi
tations
of
e
xisting
research,
this
paper
proposes
the
dynamic
semantic
fog
(DSF)
frame
w
ork,
which
performs
semantic
processing
directly
at
the
fog
layer
to
achie
v
e
real-time
semantic
interop-
erability
and
inte
grates
four
core
functions
into
a
unied
fog-based
architecture,
setting
it
apart
f
rom
e
xisting
similar
solutions.
The
no
v
elty
of
this
w
ork
can
be
summarized
as
follo
ws:
−
This
paper
proposes
a
semantic
frame
w
ork
for
delay-sensiti
v
e
IoT
systems,
which
mo
v
es
the
full
semantic
processing
w
orko
w—co
v
ering
RDF
generation
and
ontology
management—from
cloud
serv
ers
to
fog
nodes,
and
can
signicantly
reduce
service
latenc
y
and
netw
ork
o
v
erhead.
−
Compared
with
traditional
static
ontology
systems,
the
dynamic
ontology
enhancement
mechanism
proposed
in
this
paper
can
automatically
inte
grate
ne
w
elements
during
operation,
and
requires
no
human
interv
ention
throughout
the
entire
process.
−
The
DSF
proposed
in
this
paper
inte
grates
four
core
modules:
w
orkload
of
oading,
semantic
represen-
tation,
RDF
graph
construction,
and
ontology
e
v
olution,
to
achie
v
e
scalable
and
self-e
v
olving
semantic
interoperability
.
−
As
v
eried
through
e
xperiments,
the
ne
w
service
mechanism
proposed
in
this
paper
deli
v
ers
a
50%
reduction
in
service
delay
compared
to
tradit
ional
cloud-based
semantic
interoperability
solutions,
while
also
impro
vi
ng
ener
gy
ef
cienc
y
and
reducing
netw
ork
utilization.
The
remainder
of
this
paper
is
or
g
anized
as
follo
ws.
Section
2
re
vie
ws
e
xisting
w
ork
on
semant
ic
interoperability
and
fog-assisted
IoT
frame
w
orks.
Section
3
sho
ws
system
model
and
research
problem
of
this
w
ork.
Section
4
presents
the
proposed
fog-based
adapti
v
e
semantic
mechanism
and
describes
its
architectural
components
in
detail.
Section
5
discusses
the
e
xperimental
setup,
e
v
aluation
methodology
,
the
performance
results
and
comparati
v
e
analysis.
Finally
,
section
6
concludes
the
paper
and
outlines
directions
for
future
research.
2.
RELA
TED
W
ORK
Interoperability
is
a
well
kno
wn
issue
of
IoT
systems
because
of
the
heterogeneity
present
in
de
vices,
communication
protocols,
data
formats
and
application
domains.
In
response
to
this
challenge,
se
v
eral
researches
ha
v
e
e
xplored
interoperability
under
the
terms
de
vice,
netw
ork,
syntactic,
platform,
and
semantic
interoperability
.
Of
these
dimensions,
semantic
interoperability
stands
out
as
the
most
important
one
since
it
not
only
allo
ws
data
e
xchange
b
ut
also
f
acilitates
a
mutual
understanding
of
the
semantics
of
data
among
heterogeneous
IoT
artif
acts.
The
primary
ef
f
orts
for
semantic
interoperability
in
IoT
are
based
on
ontology-based
methods
that
represent
domain
kno
wledge
using
semantic
technologies
(such
as
RDF
and
O
WL)
[
20
]–[
22
].
These
methods
enable
an
uniform
and
machine-readable
representation
of
IoT
data
which
enables
semantic
reasoning
and
Int
J
Recongurable
&
Embedded
Syst,
V
ol.
15,
No.
3,
No
v
ember
2026:
819–836
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Recongurable
&
Embedded
Syst
ISSN:
2089-4864
❒
821
inter
-domain
data
inte
gration.
Ho
we
v
er
,
the
maj
ority
of
ontology-based
approaches
consider
a
static
en
vironment
in
which
the
de
vice
types
are
pre
dened
and
constant
ontological
structures
are
used
[
23
]–[
25
].
The
y
are
thus
not
easily
amenable
to
dynamic
IoT
scenarios
where
ne
w
de
vices,
data
attrib
utes
and
services
are
constantly
added,
calling
for
cumbersome
ontology
update
and
reconguration
[26],
[27].
In
order
to
cope
with
scalability
and
processing
issues
of
IoT
de
vices,
a
number
of
w
orks
ha
v
e
been
proposed
which
use
cloud-based
semantic
interoperability
fra
me
w
orks
[
28
].
In
the
prior
art
systems,
IoT
data
is
sent
to
centralized
cloud
serv
ers
for
both
semantic
annotati
on
and
reasoning
services
as
well
as
interoperability
[
29
]–[
30
].
Although
cloud-based
systems
pro
vide
plentiful
computing
capacity
and
gl
o
ba
l
kno
wledge
retention,
the
y
come
at
the
cost
of
serious
constraints
for
real-time
IoT
applications
[
31
],
[
32
].
In
particular
,
the
ra
w
data
is
transmitted
to
the
cloud
continuously
whic
h
incurs
higher
service
delay
,
e
xcessi
v
e
netw
ork
traf
c
and
more
ener
gy
consumption
and
therefore
is
not
viable
for
applications
with
lo
w
latenc
y
requir
ement
lik
e
smart
transportation,
smart
healthcare
auditory
IoT
,
and
industrial
IoT
.
A
more
recent
paradigm,
fog
computing,
aims
at
o
v
ercoming
some
of
these
dra
wbacks
and
of
f
ers
a
platform
between
the
IoT
de
vices
and
the
cloud
[
26
],
[
32
].
F
og-aided
semantic
interoperability
frame
w
orks
also
push
some
of
the
semantic
processing
to
w
ards
data
sources
by
deplo
ying
semantic
services
at
fog
nodes
[
33
],
[
34
].
The
current
fog-based
solutions
sho
w
better
performance
in
latenc
y
reduction
and
bandwidth
utili
zation
than
cloud-alone
based
approach.
Ho
we
v
er
,
man
y
of
these
solutions
are
still
b
uilt
on
static
ontologies
or
partial
semantic
reasoning
at
the
fog
layer
,
with
intensi
v
e
reasoning
and
ontology
e
v
olution
is
cloud-centric.
Their
payof
f
with
full
dynamic
semantic
interoperation
is
therefore
v
ery
constrained
[35]–[37].
Moreo
v
er
,
some
research
use
s
semantic
netw
orks
or
h
ybrid
semantic
approaches
to
model
IoT
kno
wl-
edge.
Although
the
y
are
easy
for
human
visualization
and
hierarchical
kno
wledge
representation,
semantic
netw
orks
do
not
ha
v
e
formal
semantics
and
reasoning
support
as
ontology-based
methods
[
38
]–[
40
].
This
limitation
therefore
limits
the
applicability
of
DCRAPs
in
lar
ge-scale
and
heterogeneous
IoT
scenarios
where
consistent
interpretation
and
e
xtensibility
are
important
[41],
[42].
Building
on
the
rele
v
ant
discussions
in
the
preceding
te
xt,
this
study
identies
three
core
research
g
aps
in
the
eld
of
IoT
semantic
processing:
t
he
lack
of
mechanisms
to
dynamically
inte
grate
ne
w
IoT
de
vices
at
runtime,
e
xcessi
v
e
reliance
on
cloud
processing
that
triggers
increases
in
latenc
y
,
netw
ork
o
v
erhead
and
ener
gy
consumption,
and
the
f
ailure
to
le
v
erage
fog
computing
resources
to
support
adapti
v
e
semantic
reasoning.
There
remain
unaddressed
research
g
aps
in
the
heterogeneous
IoT
domain.
There
is
an
ur
gent
need
for
a
fog-based
semantic
frame
w
ork
equipped
with
the
capabilities
of
local
semantic
processing,
dynamic
ontology
e
xpansion,
and
runtime
semantic
adaptation.
This
paper
proposes
a
fog-based
adapti
v
e
semantic
mechanism
deplo
yed
at
the
fog
layer
,
which
inte
grates
four
core
modules
including
w
orkload
of
oading,
RDF
generation,
and
semantic
representation.
Its
core
goal
is
to
achie
v
e
scalable,
ener
gy-ef
ci
ent
r
eal-time
semantic
interoperability
for
heterogeneous
IoT
systems.
T
able
1
summarizes
the
dif
ferences
between
representati
v
e
semantic
interoperability
approaches
and
the
proposed
DSF
frame
w
ork.
T
able
1.
Comparison
of
e
xisting
semantic
interoperability
frame
w
orks
and
the
proposed
DSF
Frame
w
ork
F
og-based
RDF
at
fog
Dynamic
ontology
Runtime
de
vice
inte
gration
Ontology-based
methods
×
×
×
×
Cloud-based
semantic
frame
w
orks
×
×
×
P
artial
F
og
semantic
web
frame
w
orks
✓
P
artial
×
P
artial
Dyno-IoT
[23]
×
×
✓
P
artial
Pr
oposed
DSF
✓
✓
✓
✓
3.
SYSTEM
MODEL
AND
PR
OBLEM
DEFINITION
This
section
presents
the
system
model
under
study
and
formulates
the
problem
of
dynamic
semantic
interoperability
in
heterogeneous
IoT
.
The
proposed
model
sets
the
assumptions,
architectural
background
and
design
considerations
to
which
the
fog-based
adapti
v
e
semantic
mechanism
refers.
3.1.
System
model
The
system
model
used
in
this
paper
is
presented
as
a
series
of
logical
steps
that
indicate
ho
w
disparate
IoT
data
are
generated,
processed
and
semantically
interpreted
within
the
en
visioned
IoT
–fog–cloud
model,
as
sho
wn
in
Figure
1.
Heter
o
g
eneous
internet
of
things
de
vice
inter
oper
ability
using
fo
g-based
...
(De
vamekalai
Na
gasundar
am)
Evaluation Warning : The document was created with Spire.PDF for Python.
822
❒
ISSN:
2089-4864
−
Heterogeneous
IoT
data
generation:
thousands
of
not
similar
IoT
de
vices,
such
as
temperature
sensors
and
actuators,
are
scattered
in
the
tar
get
en
vironment
(e.g.,
smart
city
use
cases).
These
de
vices
produce
continuous
streams
of
data
that
emplo
y
v
arious
formats,
names,
units
and
schemas.
−
Data
transmission
to
fog
nodes:
the
IoT
de
vices
send
their
data
to
t
he
nearest
fog
node
o
v
er
their
def
ault
communication
infrastructure.
The
basic
netw
orking
interoperation
is
presumed
to
already
e
xist
that
permits
de
vices
to
communicate
without
conicts
in
a
protocol-le
v
el
conte
xt.
−
F
og
node
reception
and
resource
a
v
ailabili
ty:
each
fog
node
has
limited
computational
and
storage
capacity
,
serving
as
middle
processing
tier
between
IoT
de
vices
and
cloud.
F
og
nodes
are
constantly
tracking
their
resource
utilization
and
service
queue
lengths,
in
terms
of
both
CPU
usage,
a
v
ailable
memory
,
and
task
queues.
−
Local
data
processing
at
the
fog
tab:
when
recei
ving
the
data,
fog
nodes
locally
process
(i.e.,
pack
etize,
transform,
lter
,
and
e
xamine)
them.
This
process
operates
as
a
lter
to
normalise
and
v
alidate
ra
w
data
which
is
of
di
v
erse
qualitati
v
e
character
prior
to
semantic
processing.
−
Semantic
encoding
and
ontology
interaction:
transformed
data
into
RDF-based
semantic
representations
and
link
ed
with
the
local
ontology
.
In
case
the
incoming
data
cannot
be
entirely
mapped
into
a
v
ailable
ontology
concept,
the
dynamic
enrichment
process
will
be
initiated
to
dynamically
update
semantic
model.
−
Selecti
v
e
cloud
interaction:
only
semantically
digested,
annotated,
or
non-time-sensiti
v
e
data
are
transmitted
to
the
cloud
tier
for
archi
v
al
or
w
orldwide
analytics.
Latenc
y-sensiti
v
e
semantic
interoperability
tasks
are
k
ept
at
the
fog
layer
for
real-time
responsi
v
eness.
−
Continuous
system
e
v
olution:
with
the
ne
w
de
vices
and
types
of
data,
the
system
can
update
its
ontologies
and
semantic
mappings
in
runtime
completely
automatic
which
allo
ws
to
pro
vide
a
scalable
and
dynamic
semantic
interoperability
without
requiring
manual
reconguration
of
all
de
vices.
Figure
1.
Ov
erall
system
model
of
the
proposed
fog-based
adapti
v
e
semantic
interoperability
mechanism
The
proposed
architecture
f
acili
tates
agile
semantic
interoperability
by
partitioning
the
processing
responsibilities
across
IoT
,
fog,
and
cloud
layers
in
an
hierarchical
f
ashion
emphasizing
lo
w
latenc
y
,
scalability
,
and
recongurability
in
heterogeneous
IoT
platforms.
3.2.
Pr
oblem
denition
Although
it
is
being
widely
applied
in
IoT
systems
lar
ge
scale
semantic
technologies,
current
solutions
to
semantic
interoperability
ha
v
e
some
weaknesses
particularly
when
considering
real-time
and
lar
ge
and
Int
J
Recongurable
&
Embedded
Syst,
V
ol.
15,
No.
3,
No
v
ember
2026:
819–836
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Recongurable
&
Embedded
Syst
ISSN:
2089-4864
❒
823
heterogeneous
scenarios.
These
problems
may
be
formulated
as
follo
ws:
−
High
service
delay:
the
e
xplanation
is
that
cloud-based
semantic
interoperability
solutions
implies
sending
data
to
remote
serv
ers
for
sem
antic-related
processing,
with
a
consequent
high
latenc
y
inadequate
for
real-time
IoT
applications.
−
Excessi
v
e
netw
ork
usage:
when
ra
w
or
minimally
processed
IoT
data
are
sent
to
the
cloud,
high
bandwidth
usage
and
netw
ork
o
v
erload
can
result,
allo
wing
for
deplo
yment
scenarios
featuring
man
y
de
vices.
−
High
ener
gy
consumption:
frequent
long
range
wireless
communication
and
central
semantic
computation
lead
to
high
ener
gy
consumption,
which
is
not
sustainable
for
IoT
systems.
−
No
dynamic
semantic
adaptability:
other
approaches
use
x
ed
ontologi
es
for
which
manual
changes
are
required
as
ne
w
de
vices
and
data
types
emer
ge,
making
the
model
unt
for
e
v
olving
IoT
ecosystems
with
dynamic
contents.
Ag
ainst
this
background,
the
problem
being
considered
in
this
study
can
be
formulated
as
follo
ws:
ho
w
to
realize
dynamic
semantic
interoperability
in
heterogeneous
IoT
systems
with
the
requirement
of
ef
cient
delay
,
ef
fecti
v
e
lessening
of
netw
ork
o
v
erhead
and
po
wer
consumption
as
well
as
adapti
v
e
inte
gration
of
ne
w
de
vices
and
semantics
simultaneously?
T
o
this
end,
the
proposed
method
utilises
fog
computing
to
push
do
wn
semantic
processing
into
data
sources
and
introduces
an
adapti
v
e
ontology
enrichment
capability
that
enables
a
dynamic
e
v
olution
of
the
semantic
model.
This
fusion
all
o
w
performance
ef
fecti
v
e,
scalable,
and
real-time
semantic
interoperability
in
Heterogeneous
IoT
systems.
4.
PR
OPOSED
FOG-B
ASED
AD
APTIVE
SEMANTIC
MECHANISM
In
this
section,
we
introduce
our
proposed
fog-based
adapti
v
e
semantic
mechanism
called
the
DSF
for
supporting
runtime
and
dynamic
interoperability
among
heterogeneous
IoT
de
vices
t
o
alle
viate
service
delay
,
netw
ork
utilization
cost
and
ener
gy
consumption
by
of
oading
ontology
processing
from
resource-constrained
de
vices
to
their
neighboring
fog
nodes.
The
DSF
model
in
v
olv
es:
i)
w
orkload
shifting
in
fog
layers,
ii)
data
getting
in
appropriate
formats
and
semanti
c
representation,
iii)
creation
of
RDF
graph
database,
and
i
v)
e
xible
ontology
e
xtension
for
the
automatic
inte
gration
of
ne
w
de
vices
or
data
sources.
4.1.
Ov
erall
ar
chitectur
e
The
fog-based
adapt
i
v
e
semantic
mechanism
is
al
so
composed
of
three
layers
(IoT
layer
,
fog
layer
,
and
cloud
layer)
to
dri
v
e
dynamic
implementation
of
semantic
interoperability
in
IoT
h
ybrid
en
vironments,
as
sho
wn
in
Figure
2.
This
design
is
based
on
the
f
act
that
semantic
processing
is
placed
in
the
fog
layer
to
eliminate
such
shortcomings
of
cloud-centric
architect
ures
as
a
trick
y
o
v
erhead
associated
with
high
response
time
and
e
xcessi
v
e
netw
ork
usage
and
ener
gy
consumption
in
real-time
IoT
applications.
−
IoT
layer:
heterogeneous
de
vices
s
uch
as
sensors,
actuators
and
smart
objects
produce
data
in
v
arious
formats
(e.g.,
JSON,
XML,
and
CSV),
b
ut
also
semantic
represe
ntations.
W
e
assume
that
these
de
vices
ha
v
e
already
established
basic
netw
orking
interoperability
,
and
t
he
y
mainly
concentrate
on
sensing
data
and
transmitting
the
data
with
less
local
processing
resulting
from
resource
limitations.
−
The
fog
layer:
is
the
heart
of
the
presented
mechanism.
F
og
nodes
are
located
near
the
IoT
de
vices
and
ha
v
e
moderate
computing
po
wer
and
storage.
This
layer
in
v
olv
es
semantic
processing
modules:
w
orkload
of
oading,
data
fetching
and
interpretation,
generation
of
the
RDF
representation
and
adapti
v
e
ontology
enrichment.
By
performing
these
tasks
locally
,
the
fog
layer
enables
lo
w-latenc
y
semantic
inference,
reduces
traf
c
to
cloud
services
and
allo
ws
for
easy
inclusion
of
ne
w
de
vices
or
data
modalities
into
the
semantic
model
in
a
dynamic
w
ay
.
−
Cloud
layer:
the
cloud
layer
is
an
auxiliary
component
in
char
ge
of
data
storage,
historical
analysis
and
global
systems
management.
On-demand
non-time-sensiti
v
e
tasks
or
compressed
semantic
summaries
can
then
be
sent
to
the
cloud.
Ho
we
v
er
,
the
architecture
is
designed
to
be
independent
of
cloud
as
much
as
possible
for
a
real-time
semantic
interoperation
to
pre
v
ent
potential
performance
bottleneck.
The
o
v
erall
layered
architecture
ensures
that
the
semantic
int
eroperability
is
dynamically
and
ef
c
iently
realized
in
an
intelligence
dispersed
manner
o
v
er
IoT–fog–cloud
continuum.
The
fog-dri
v
en
approach
supports
dynamic
scalability
,
enhances
the
responsi
v
eness
for
real-time
services
and
yields
a
persistent
solution
to
cope
with
semantic
heterogeneity
in
lar
ge-scale
IoT
systems.
Heter
o
g
eneous
internet
of
things
de
vice
inter
oper
ability
using
fo
g-based
...
(De
vamekalai
Na
gasundar
am)
Evaluation Warning : The document was created with Spire.PDF for Python.
824
❒
ISSN:
2089-4864
Figure
2.
Three-layer
IoT
-fog-cloud
architecture
of
the
proposed
fog-based
adapti
v
e
semantic
mechanism
[
26
]
4.2.
Node
w
orkload
ofoading
module
The
fog
node
w
orkload
of
oading
module
is
a
k
e
y
b
uilding
block
of
the
proposed
fog-based
adapti
v
e
semantic
mechanism,
in
char
ge
of
e
x
ecuting
semantic
processing
tasks
so
that
real-time
requirements
are
met
and
resource
utilization
is
optimized,
as
sho
wn
in
Figure
3.
Because
it
is
computationally
intensi
v
e
in
ontological
reasoning
and
RDF
generation,
to
mak
e
all
semantic
logic
operations
occurring
at
the
cloud
side
will
result
in
service
delay
and
o
v
erhead.
T
o
o
v
ercome
this
challenge,
the
proposed
system
prefers
to
perform
fog-side
processing
for
latenc
y-sensiti
v
e
tasks
and
can
dynamically
upload
unnecessary
data
to
cloud
when
possible.
Figure
3.
Design
of
the
fog
node
w
orkload
of
oading
module
for
adapti
v
e
semantic
processing
The
of
oading
module
w
orks
in
a
round-robin
f
ashion
w
atches
the
resource
status
for
fog
nodes
and
cate
gorizes
incoming
semantic
w
orkloads,
depending
on
their
ur
genc
y
and
computational
needs.
Operations,
such
as
data
parsing
and
ltering,
R
DF
formation
and
ontology
updat
e
are
considered
of
high
priority
and
the
y
are
performed
locally
at
the
fog
le
v
el
for
immediate
semantic
interpretation.
Ho
we
v
er
,
tasks
that
are
demandingly
computationally
hea
vy
or
non-real-time
ones
(e.g.,
lar
ge-scale
historical
reasoning
and
long-term
analytics)
can
be
pushed
to
the
cloud
layer
in
order
not
to
o
v
erload
resource
within
a
fog.
The
fog
node
w
orkload
of
oading
module
comprises:
−
W
orkload
discriminator:
recognizes
dif
ferent
semantic
processing
tasks
and
assigns
them
a
real-time
require-
ment
and
computation
comple
xity
based
on
their
nature.
−
Resource
monitor:
monitors
fog
node
resourc
es
(CPU
usage,
memory
a
v
ailability
,
and
task
queue
length)
at
e
v
ery
instant
of
time.
−
Of
oading
decision
engine:
decides
whether
a
task
is
best
e
x
ecuted
locally
or
of
oaded
to
the
cloud
by
considering
a
v
ailable
resources
and
quality
of
service
constraints.
−
Ex
ecution
scheduler:
the
e
x
ecution
scheduler
or
g
ani
zes
to
run
accepted
tasks
ef
ciently
and
within
the
deadline.
By
introducing
intelligent
task
placement
and
adapti
v
e
of
oading,
this
module
can
greatly
decreas
e
the
Int
J
Recongurable
&
Embedded
Syst,
V
ol.
15,
No.
3,
No
v
ember
2026:
819–836
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Recongurable
&
Embedded
Syst
ISSN:
2089-4864
❒
825
service
delay
and
netw
ork
traf
c-consuming
meanwhile
k
eeping
high
semantic
processing
accurac
y
.
Addition-
ally
,
the
of
oading
strate
gy
from
fog-based
systems
i
mpro
v
es
its
scalability
by
a
v
oiding
an
y
cloud
bottleneck
and
additionally
f
acilitates
the
simple
addition
of
ne
w
IoT
de
vices
and
semantic
w
orkloads
to
the
system
as
it
gro
ws.
4.3.
Data
pr
ocur
ement
and
r
epr
esentation
The
data
procurement
and
representation
is
in
char
ge
of
collecting
data
from
the
heterogeneous
IoT
streams,
which
is
then
put
into
homogeneous
semantically
mea
ningful
form
to
be
used
for
follo
wing
steps
including
RDF
generation
and
ontology
enrichment,
as
sho
wn
in
Figure
4.
Ne
w
da
ta
may
be
of
dif
f
erent
syntax,
format,
measurement
units
and
le
v
el
of
detail
according
to
the
lar
ge
num
b
e
r
of
IoT
pro
viders
and
type
of
de
vices.
This
layer
allo
ws
for
ra
w
de
vice
stream
information
to
be
normalized,
v
a
lidated
and
ltere
d
before
it
is
processed
semantically
on
the
fog,
minimising
redundant
computation
and
netw
ork
traf
c
to
w
ards
cloud.
Figure
4.
Design
of
the
data
procurement
and
representation
module
for
heterogeneous
IoT
data
processing
This
module,
as
depicted
throughout
the
paper
,
consists
of
three
consecuti
v
e
stages:
data
parsing
and
transformation,
data
ltration,
and
data
inspection.
The
combi
nation
of
all
these
steps
allo
ws
a
highly
ef
fecti
v
e
processing
of
heterogeneous
data
in
the
sense
that
semantics
is
preserv
ed.
−
Data
parser
and
t
ransformation:
the
data
parser
detects
the
syntactic
structures
of
input
records
(i.e.,
JSON,
XML,
or
CSV),
and
prunes
irrele
v
ant
elds.
The
transformation
f
actor
subsequently
describes
these
attrib
utes
in
a
normalized
internal
representation
of
the
name
con
v
e
ntions,
data
types,
and
measuring
units.
This
stage
gi
v
es
an
homogeneous
structure
of
the
data,
so
that
is
easier
to
perform
semantic
mapping
and
RDF
generation.
−
Data
ltration:
the
reduction
stage
lters
out
redundant,
incomplete,
and
irrele
v
ant
data
records
to
reduce
the
processing
o
v
erhead
and
netw
ork
ut
ilisation.
Filtering
rules
could
check
for
duplicate
remo
v
al,
v
alue-based
thresholds
and/or
application-specic
attrib
ute
pruning.
Through
early
dropping
of
unnecessary
information,
the
proposed
scheme
can
ef
ciently
support
real-time
IoT
applications
with
lo
w
latenc
y
.
−
Data
inspection:
the
data
re
vie
w
process
checks
normalized
records
for
consistenc
y
and
completeness
before
semantic
encoding.
It
tests
for
the
presenc
e
of
required
elements,
v
alidates
range,
and
forms
structure.
This
ensures
that
i
ncorrect
or
inconsistent
data
do
not
propag
ate
into
the
semantic
layer
a
nd
ultimately
impro
v
es
the
reliability
of
ontology
enrichment
and
reasoning
operations.
As
a
result,
the
data
procurement
and
representation
module
is
mainly
responsible
to
act
as
an
interf
ace
between
ra
w
poorly-structure
d
IoT
data
forms
and
the
semantic
interoperability
space
by
sharing
clean,
normalised
and
sem
antically
meaningful
data
to
pass
for
RDF
generation
and
ontology
processing
on
fog
layer
.
4.4.
Resour
ce
description
framew
ork
generation
and
semantic
r
epr
esentation
Con
v
erts
IoT
data
into
machine
underst
andable
semantic
format,
RDF
.
RDF
is
a
standard
and
e
x-
tensible
data
model
that
can
achie
v
e
semantic
i
nteroperability
by
formatting
information
in
the
triples
of
subject–predicate–object,
as
sho
wn
in
Figure
5.
Such
a
representation
entails
that
heterogeneous
IoT
de
vices
and
services
communicate
with
shared
meaning,
re
g
ardless
of
natural
metadata
formats
or
manuf
acturer
.
In
this
module,
after
normalization
and
v
alidation
of
the
data
processed
in
the
pre
vious
module,
the
fog
node
starts
semantic
encoding
by
capturing
conte
xtual
meta
data
and
mapping
de
vice
observ
ations
to
concepts
within
the
ontology
.
Performing
RDF
generation
in
the
fog
layer
can
bring
do
wn
semantic
proces
sing
time
and
reduce
ra
w
data
transmission
to
the
cloud.
The
procedure
to
deri
v
e
the
RDF
in
v
olv
es
a
number
of
stages.
−
Metadata
e
xtraction:
the
e
xtracted
metadata
contained,
including
de
vice
IDs,
timestamps,
location
informa-
tion,
follo
wed
by
measurement
units
and
conte
xtual
tags.
Heter
o
g
eneous
internet
of
things
de
vice
inter
oper
ability
using
fo
g-based
...
(De
vamekalai
Na
gasundar
am)
Evaluation Warning : The document was created with Spire.PDF for Python.
826
❒
ISSN:
2089-4864
−
Metadata
annotation:
it
refers
to
relating
of
e
xtracted
metadata
with
ontology
classes
and
properties
by
means
of
predened
v
ocab
ularies
and
semantic
rules.
This
action
pro
vides
uniform
interpretation
of
data
in
di
v
erse
sources.
−
RDF-triples
construction:
for
each
v
alidated
data
observ
ation,
we
encode
i
t
into
an
RDF
triples
structure,
linking
the
entities
of
de
vices,
their
observ
ed
v
alues,
and
conte
xt
attrib
utes
in
a
graph
form.
The
produced
RDF
graph
creates
a
homogeneous
semantic
layer
for
supporting
semantic
querying,
reasoning
and
interoperability
servi
ces.
Using
a
unied
semantic
representation
for
IoT
data,
the
proposed
mid-
dle
w
are
supports
coherent
inte
gration
of
ne
w
de
vices
and
data
types
without
an
y
need
of
manual
reconguration.
In
addition,
the
RDF
based
semanti
c
representa
tion
enables
scalabi
lity
and
e
xtensibility
as
ne
w
semantic
relations
can
be
added
dynamically
by
ontology
enrichment.
This
creates
a
common
semantic
understanding
of
things
across
IoT–fog–cloud
continuum
and
this
formulation
forms
the
base
for
dynamic
semantic
interoperability
in
our
system.
Figure
5.
Softw
are
architecture
of
the
RDF
generation
and
semantic
representation
module
at
the
fog
layer
4.5.
Adapti
v
e
ontology
enrichment
The
adapt
i
v
e
ontology
enrichment
is
the
crux
of
the
proposed
fog-based
adapti
v
e
semantic
mechanism,
allo
wing
for
dynamic
semantic
interoperability
in
heterogeneous
IoT
en
vironments.
In
contrast
to
static
semantic
approaches
based
on
a
priori
and
x
ed
ontologies,
this
module
permits
the
ontology
to
nati
v
ely
gro
w
at
runtime
by
assimilating
ne
w
de
vices,
attrib
utes
and
relationships
as
the
y
appear
,
as
sho
wn
in
Figure
6.
By
performing
the
ontology
enrichment
in
fog,
proposed
approach
enables
real-time
adaptabilit
y
with
no
latenc
y
and
o
v
erhead
requirements
of
cloud-based
ontology
management.
The
enrichment
process
is
initial
ised
when
the
incoming
IoT
data
cannot
be
completely
related
with
e
xisting
ontology
classes
or
properties.
In
such
cases,
the
fog
node
autonomously
process
data
structure
and
semantics,
update
the
ontology
appropriately
,
and
distrib
utes
such
ne
w
kno
wledge
in
order
to
potentially
ha
v
e
it
reused
later
on.
This
capacity
allo
ws
for
the
inte
gration
of
ne
w
de
vices
and
data
types
without
manual
ef
fort.
The
enrichment
is
acti
v
ated
whene
v
er
there
are
incoming
IoT
messages
that
cannot
be
correctly
annotated
with
an
e
xisting
ontology
class
or
property
.
In
this
situation,
the
fog
node
processes
on
its
o
wn
the
structure
and
meaning
of
the
data
it
recei
v
es,
updating
its
ontology
accordingly
,
and
sa
ving
ne
w
kno
wledge
to
reuse
in
later
occasions.
This
ability
allo
ws
for
the
automatic
inclusion
of
ne
w
or
unkno
wn
de
vices
and
data
types
with
minimal
human
interv
ention.
−
Attrib
ute
mapping:
attrib
ute
mapping
matches
incoming
data
elds
to
ontology
attrib
utes.
Kno
wn
at
trib
utes
are
straightforw
ardly
mapped
to
kno
wn
terms
while
the
unkno
wn
ones
are
analyzed
in
the
light
of
conte
xt
and
string
simi
larity
for
ndi
ng
appropriate
sem
antic
representations.
When
v
alidated,
ne
w
attrib
utes
mapping
is
re
gistered
to
ha
v
e
dra
wers
interpret
future
streams
reliably
.
−
Primary
subject
identication:
this
stage
determines
the
main
semant
ic
subject
of
arri
ving
data,
for
e
xample
a
particular
sensor
,
de
vice
or
monitored
object.
Accurate
subject
identication
guarantees
that
the
ne
wly
Int
J
Recongurable
&
Embedded
Syst,
V
ol.
15,
No.
3,
No
v
ember
2026:
819–836
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Recongurable
&
Embedded
Syst
ISSN:
2089-4864
❒
827
generated
RDF
triples
are
connected
to
the
right
ontology
instances
so
as
not
to
disturb
semantic
coherence
throughout
the
kno
wledge
graph
(KG).
−
Ontology
class
assessment
and
identi
cation:
for
ne
w
de
vice
types
or
application
domains,
it
is
e
xamined
whether
an
appropriate
ontological
class
already
e
xists
for
data
pro
vided.
Whe
n
no
appropriate
class
is
retrie
v
ed,
we
create
a
ne
w
class
and
insert
it
in
the
ontol
o
gy
by
creating
subclass
relationships.
Appropriate
features
and
constraints
are
then
assigned
to
the
ne
w
class
gi
v
en
observ
ed
data
attrib
utes.
−
Mapping
representation
and
class
v
erication:
the
v
eried
attrib
ute
and
class
mappings
are
formally
articu-
lated
as
ontology-grounded
statements.
The
v
erication
step
ensuring
the
consistenc
y
and
correctness
of
ne
wly
added
classes,
properties
with
respect
to
the
ontology
constraints
such
as
domain/range
restrictions.
−
Ontology
enrichment
and
update:
nally
,
the
ontology
is
e
xtended
with
the
ne
wly
v
alidated
classes,
properties
and
relations.
This
information
w
ould
allo
w
the
system
to
understand
semantically
subsequent
data
coming
from
an
y
ne
w
inserted
de
vice
automatically
.
Such
an
adapti
v
e
enrichment
procedure
allo
ws
to
be
scalable,
v
endor
independent
and
e
v
olv
e
the
IoT
ecosystem,
seamlessly
.
Figure
6.
Process
o
w
of
the
adapti
v
e
ontology
enrichment
module
enabling
dynamic
semantic
interoperability
at
the
fog
layer
In
this
re
g
ard,
the
proposed
mechanism
can
achie
v
e
real-time
semantic
interoperability
in
a
true
dynamic
manner
through
its
autonomous
ontology
e
v
olution
at
fog
layer
which
is
a
mandatory
requirement
for
critical
and
lar
ge-scale
deplo
yments
of
IoT
.
Heter
o
g
eneous
internet
of
things
de
vice
inter
oper
ability
using
fo
g-based
...
(De
vamekalai
Na
gasundar
am)
Evaluation Warning : The document was created with Spire.PDF for Python.
828
❒
ISSN:
2089-4864
5.
RESUL
TS
AND
PERFORMANCE
AN
AL
YSIS
In
this
section,
we
introduce
the
e
xperimental
results
and
performance
analysis
on
fog-based
adapti
v
e
semantic
mechanism.
The
e
v
aluation
aims
at
judging
whether
the
performance
concerning
the
semantic
interoperability
performs
an
y
better
when
it
comes
to
implementing
that
in
fog
layer
instead
of
re
gular
cloud-
based
semantic
solutions.
Performance
is
objecti
v
ely
e
v
aluated
through
three
metrics
–
service
delay
,
ener
gy
consumption
and,
the
netw
ork
usage.
Such
metrics
are
particularly
important
for
real-time
and/or
massi
v
e-scale
IoT
applications
that
focus
on
demands
of
lo
w
latenc
y
,
resource
usage
ef
cienc
y
,
and
scalability
.
5.1.
Experimental
setup
The
ef
f
ecti
v
eness
of
the
proposed
fog-based
adapti
v
e
semantic
mechanism
w
as
tested
with
a
simulation-
dri
v
en
e
xperimental
platform
to
w
ork
under
e
v
aluation
in
relation
with
enabling
dynamic
IoT
en
vironments
for
hetere
geneous
sementic
interoperability
.
The
e
xperiments
were
performed
based
on
the
iF
ogSim
simulator
,
a
popular
toolkit
to
simulate
and
analyze
fog
and
edge
computing
en
vironments,
as
sho
wn
in
Figure
7.
A
smart
city
traf
c
management
scenario
w
as
chosen
as
an
instance
of
a
real
and
data
intensi
v
e
w
orkload
in
the
IoT
,
including
de
vices
with
heterogeneity
,
real-time
data
streams,
and
latenc
y
sensiti
v
e
a
pp
l
ications.
F
og
nodes
were
located
near
to
IoT
de
vices
and
process
semantic
tasks,
whereas
the
cloud
layer
w
as
utilized
for
non-real-time
analytics
as
well
as
long-term
storage.
Figure
7.
DSF
implementation
simulation
topology
F
or
a
f
air
and
replicable
e
v
aluation,
the
simulation
parameters,
hardw
are
settings,
and
softw
are
en
vironment
were
preserv
ed
during
all
e
xperiments.
The
performance
of
the
proposed
mechanism
w
as
assessed
in
terms
of
service
delay
,
netw
ork
utilization,
and
ener
gy
usage.
These
st
atistics
sho
w
the
benet
of
ha
ving
semantic
processing
being
done
in
the
fog
as
opposed
to
a
cloud-centric
method.
The
full
hardw
are
and
softw
are
specications
of
the
e
xperimental
platform
are
sho
wn
in
T
able
2.
T
able
2.
Hardw
are
and
softw
are
specication
of
the
e
xperimental
setup
Component
Specication
Simulation
tool
iF
ogSim
Programming
language
Ja
v
a
IoT
application
scenario
Smart
city
traf
c
management
F
og
node
CPU
Multi-core
processor
(GHz
range)
F
og
node
memory
Congured
for
semantic
processing
w
orkloads
Operating
system
Linux-based
en
vironment
Data
formats
JSON,
XML,
and
CSV
Ev
aluation
metrics
Service
delay
,
netw
ork
usage,
and
ener
gy
consumption
5.2.
Ser
vice
delay
analysis
Service
delay
is
ho
w
long
IoT
data,
from
the
time
data
are
generated
to
a
de
vice
until
a
sema
ntic
response
being
issued,
is
processed
end-to-end.
This
measure
consists
of
communication
latenc
y
,
semantic
processing
delay
,
and
decision-making
time,
it
has
amen
in
v
olv
ed
into
i
mportant
aspect
for
real-time
IoT
applications
(i.e.
intelligent
transport
management
systems).
Int
J
Recongurable
&
Embedded
Syst,
V
ol.
15,
No.
3,
No
v
ember
2026:
819–836
Evaluation Warning : The document was created with Spire.PDF for Python.