TELKOM
NIKA
, Vol.14, No
.1, March 2
0
1
6
, pp. 326~3
3
4
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v14i1.2745
326
Re
cei
v
ed O
c
t
ober 3, 20
15;
Revi
se
d Ja
n
uary 6, 2016;
Acce
pt
ed Jan
uary 30, 201
6
An Autonomic Optimization Model of Multi-Layered
Dependability for Intelligent Internet of things
Zheng Ruijuan, Zhang Mi
ngchua
n*, Wu Qingtao, Li
Ying, Wei Wang
y
a
ng, Bai Xiuling
Coll
eg
e of Information En
gi
ne
erin
g, Hena
n U
n
iversit
y
of Sci
ence a
nd T
e
chnol
og
y, Lu
o
y
an
g, Hena
n, Chi
n
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: rj
w
o
@1
63.co
m
A
b
st
r
a
ct
Accompa
n
yin
g
w
i
th the spe
e
d
in
g up
of Internet of
thi
ngs
(IOT
) construction, the d
e
p
end
abi
lity
prob
le
ms
bec
o
m
e
the
i
m
p
o
rt
ant factors
con
s
traini
ng
it
s al
l-ro
u
n
d
de
vel
o
pm
en
t. Ba
sed
on
th
e
m
u
l
t
i
-
l
e
ve
l
and
mu
ltid
ime
n
sio
nal
prop
er
ties of IOT
d
epe
nd
abi
lit
y el
ements, w
i
th the ov
er
all i
m
provi
ng of th
e
dep
en
dab
ility i
ndex
of IOT as the ulti
mate g
oal, the
de
p
e
n
dab
ility e
l
e
m
en
ts of the loc
a
l f
i
ne-tu
nin
g
i
n
e
a
c
h
layer, this pap
e
r
researches th
e chan
ge rul
e
of inte
rna
l
dep
end
abi
lity ele
m
ents in perce
pti
on lay
e
r, netw
o
rk
layer an
d busi
ness layer, an
d
a
dopts perc
eptio
n
l
a
ye
r
a
s
the
exa
m
p
l
e
,
usin
g the
method
of l
i
n
e
a
r
progr
a
m
min
g
t
o
se
ek the
b
e
s
t
prop
ortion
of
all k
i
n
d
s of
de
pen
da
bility
el
e
m
e
n
ts a
nd t
h
e
opti
m
al v
a
lu
e
s
of
the el
e
m
ents, trying to c
onstr
uct a feasi
b
le
a
u
tono
mic
opti
m
i
z
at
io
n mod
e
l f
o
r de
pen
da
bil
i
ty ele
m
e
n
ts of IOT
system
. Firstly,
accor
d
ing to the function features an
d dependability pr
operties of
eac
h layer, and c
hange
rules betw
e
e
n
the
d
e
p
end
ab
il
ity
in
dex an
d
d
epe
nd
abi
lity
e
l
ements in eac
h
l
a
yer are an
aly
z
e
d
.
Seco
n
d
ly,
base
d
o
n
th
e
dyna
mic ch
an
ges (
up
or d
o
w
n) of de
pe
nd
abil
i
ty e
l
e
m
e
n
ts in
inter
nal
e
n
viro
nment (th
a
t is
,
three lay
e
rs
i
n
IOT),
the
ratio relati
ons of
d
e
pen
da
b
ility
el
e
m
e
n
ts i
n
each
layer
are
dy
na
mic
a
lly
co
ntroll
e
d
and
adj
usted
to imple
m
ent
the loc
a
l o
p
ti
mi
z
a
t
i
o
n
, im
pr
ovin
g the ov
e
r
all a
u
ton
o
m
ic
config
uratio
n
and
auton
o
m
ic ad
j
u
sting a
b
il
ity of IOT
system. At last, examp
l
e an
alysis re
s
u
lts show
that the optimi
z
a
t
i
o
n
mo
de
l prop
ose
d
in this pa
per
can rea
l
i
z
e
th
e
substanti
a
l o
p
ti
mi
z
a
t
i
o
n
in eac
h layer of IOT
.
Ke
y
w
ords
:
IOT
, Autonomic o
p
timi
z
a
ti
on, Lin
e
r progr
a
m
min
g
, F
i
ne-tuni
ng
Copy
right
©
2016 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
The Intern
et of things (IO
T
, for sho
r
t) i
s
se
en a
s
th
e third wave of information
indust
r
y,
followin
g
co
mputer, Inte
rnet an
d mo
bi
le commu
ni
cation n
e
two
r
k. Due
to its
wide
prospe
ct of
indu
strial app
lication, IOT is bro
ught to the a
ttention of the govern
m
ents.
In the developme
n
t of
the IOT, beca
u
se entitie
s in the IOT sce
ne have
a ce
rtain perce
ption, ca
lculatio
n and execution
ability, these widespread percept
ion equipment
s will
cause new th
reat to national basi
s
, social
and
pe
rso
nal
inform
ation
secu
rity. IOT i
s
fa
cing
a
lot
of inform
ation
se
cu
rity chal
lenge
s. In
view
of the expo
sed o
r
p
o
tenti
a
l dep
end
abi
lity threat
s i
n
the pe
rcepti
on laye
r, net
work l
a
yer
a
n
d
busi
n
e
ss l
a
ye
r of IOT, fa
ci
ng the
syste
m
glob
al
an
d
overall dep
e
ndability
an
d according
to the
multidimen
sio
nal and multil
ayer archite
c
t
u
re, ho
w to b
u
ild an op
era
t
ional optimization model t
o
depe
ndability
eleme
n
ts in I
O
T that refle
c
ts gl
obal
de
pend
ability chara
c
te
risti
c
s, which be
co
mes
a key pro
b
le
m to be solve
d
urge
ntly in this field.
The esta
blish
m
ent of such
a model ha
s im
porta
nt theoretical g
u
i
dan
ce and p
r
acti
cal
signifi
cance t
o
the ascensi
on of dependability in
IOT. We
know that the
source of any net
work
system de
pe
ndability pro
b
l
em can b
e
divided into tw
o categ
o
rie
s
, external an
d internal fa
ctors.
External atta
ck, lin
k o
r
th
e device fail
ure, u
s
e
r
s m
i
sop
e
ratio
n
,
virus and oth
e
r
fa
ctors,
could
eventually lea
d
to system functio
n
de
clin
e or cra
s
h. What is impo
rta
n
t in an IOT system is that it
need
s the
wi
sdom
natu
r
e,
that is, ba
sed on th
e
d
y
namic cha
n
ges (up or d
o
wn
)
of
sy
st
e
m
depe
ndability
eleme
n
ts i
n
i
n
ternal
envi
r
o
n
ment
(that
i
s
, the sy
stem
stratifi
cation),
the
system
can
dynamically control a
nd a
d
just the
rati
o rel
a
ti
on
s of
the de
pend
ability eleme
n
ts to impl
e
m
ent
the optimi
z
ati
on in
a l
a
yer. On thi
s
b
a
sis, it
ne
ed
s t
o
coordinate
relation
s
am
ong
ea
ch l
a
yer,
redu
cin
g
the
mutual inte
rf
eren
ce
bet
we
en the
layers in the
great
est d
egree. I
n
current, m
any
literatures d
e
v
oted to anal
yzing the rea
s
on
s to net
work i
n
terrupti
on and a
bno
rmal inform
ation in
IP backbon
e
netwo
rk, Internet
servi
c
e
s
, BGP
routing, and
the results sho
w
that
the
config
uratio
n error is on
e o
f
the most important
cau
s
e
s
[1]. The configuratio
n errors have hu
ge
impact to the
depe
ndability
of the system.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
An Autonom
ic Optim
i
zatio
n
Model of M
u
lti-La
ye
red
Dep
end
ability for… (Zh
eng
Ruijua
n
)
327
As a
re
sult,
auton
omic
config
uratio
n
and man
a
g
e
ment
of co
mplex
an
d dynamic
netwo
rks is
a chall
engin
g
probl
em for netwo
rk rese
arche
r
s and de
sign
e
r
s. Such onl
ine
optimizatio
n of
system
s can
be pe
rf
ormed in t
w
o
ways: (i) u
s
ing
a se
pa
rate m
odel of the
sy
stem
for experim
e
n
ting new
co
nfiguratio
ns,
(ii) u
s
ing
the
system itself
for experim
e
n
tation witho
u
t
a
sep
a
rate
sy
stem mo
del [2
]. At the sa
me time
,
Sel
f
-org
ani
zing netwo
rk, or SON,
technol
ogy,
whi
c
h is a
b
le
to minimize h
u
man inte
rve
n
tion in networki
ng p
r
o
c
e
s
se
s, wa
s pro
posed to red
u
ce
the operation
a
l co
sts for
service p
r
ovid
ers in futu
re
wirel
e
ss sy
stems [3].
The
resea
r
ch
on the
algo
ri
thm and
mod
e
l of auto
n
o
m
ic
config
ura
t
ion and
opti
m
ization
has a
c
hieve
d
rapid dev
elopme
n
t. Referen
c
e [4]
propo
se
d an approa
ch for improvin
g
the
enterp
r
i
s
e fe
m to cell net
work’
s
pe
rformance by
a
u
tomated opt
imizing the f
e
m to cell b
a
se
station’
s (FB
S
’s) pilot po
wer a
s
well as ant
e
nna
pattern, and
the recently prop
osed mu
lti-
element ante
nna which is
approp
riate for fem to cell
is also introd
uce
d
.
On the
config
uration
an
d o
p
timization
of
auton
omou
s
mode
and
fra
m
ewo
r
k, a Mi
d-level
Network Services
Config
urati
on (MiNSC)
fram
ewo
r
k wa
s
creat
ed to overco
me the use
of
manag
eme
n
t transl
a
tion m
e
ch
ani
sms, t
o
su
ppo
rt
the
netwo
rk’
s
h
e
t
eroge
neity [5]. To guarant
ee
the config
uration man
agem
ent inter op
erability
a data model d
e
finition lang
uag
e, named YA
NG
[6], was cre
a
ted. However,
the p
r
eviou
s
pro
p
o
s
al
s
ca
nnot ma
ke
th
e integ
r
ated
manag
eme
n
t of
hetero
gen
eo
us
el
eme
n
ts while promoti
ng
thei
r
ma
n
ageme
n
t auto
m
ation. The
Mid-level
Net
w
ork
Service
s
Co
nfiguratio
n M
i
NSC frame
w
ork was
cr
ea
ted to overcome the limit
ations
of tho
s
e
transl
a
tion
mech
ani
sm
s, providing
a mid-level
manage
me
nt abstra
c
tio
n
based on
the
asso
ciation
o
f
stand
ard n
e
twork man
a
gement
interf
ace
s
and
sta
ndard n
e
two
r
k m
anag
eme
n
t
informatio
n
model
s [7].
Based
on
d
a
ta mod
e
ling
lang
uage,
a configu
r
ati
on a
u
tomatically
gene
rated
m
odel fo
r sem
antic laye
r
was d
e
si
gne
d
by YANG, El
bada
wi, etc,
whi
c
h d
e
fine
d as
the CSM. Th
e sem
antic l
a
yer can correctly,
effecti
v
ely and rea
s
on
ably de
scribe the n
e
twork
config
uratio
n
[8]. Which m
a
ke
s that the
config
uratio
n
informatio
n automatically gene
rated
can
be re
cog
n
ize
d
and parse
d
by semantic
layer, and
then be distri
b
u
ted to a spe
c
ific device. The
main ide
a
of
feature
sele
ct
ion is to
cho
o
se
a su
bset of input vari
able
s
by elim
inating featu
r
es
with little or no predi
ctive informatio
n, [9] intr
odu
ce
s two app
roa
c
h
e
s in feature sele
ction kno
w
n
as forwa
r
d se
lection a
nd b
a
ckward
sele
ction.
So far
as we kno
w
, the
r
e a
r
e a
fe
w co
nvinci
ng
studie
s
on t
he d
eployme
nt and
optimizatio
n of
so
me ce
rta
i
n
net
works, whi
c
h ca
n
b
e
fairly important in th
e future communication
system
s. But
most
re
sea
r
ches focus on
the conf
igu
r
at
ion a
nd
optim
ization
of lo
cal pe
rforman
c
e
of a sp
ecifi
c
net
work a
pplication. Consi
der
i
ng t
he multidim
ensi
onal hi
e
r
archi
c
al
system
architectu
re
of IOT syste
m
, from the
global
eyes, to drive gl
obal o
p
timization by mi
cro
adju
s
tment
and e
s
tabli
s
h an
auton
omic
optimization mo
del
in a
c
corda
n
ce
with l
a
yere
d
architectu
re i
n
IOT, has no
t been rep
o
rt
ed in literatu
r
es.
Autonomi
c
computing
im
prove
s
the
servi
c
e perfo
rman
ce
by mean
s
of a
u
tonomi
c
adju
s
tment of
softwa
r
e
and
hardwa
r
e
re
sou
r
ces,
whi
c
h gives
us im
portant
enligh
t
enment. If we
can ap
ply it to con
s
tru
c
t the depen
dabili
ty optimiz
ation model, we
maybe achie
v
e a new trai
n of
thought to implement sy
stem autono
m
y
and solve
the pro
b
lem o
f
system safe
ty performan
ce.
Many areas
have carried
out the
appli
c
ation
re
sea
r
ch of auto
n
o
m
ic comp
uting. Such a
s
t
h
e
FOCALE [10]
proje
c
t implements the a
u
tonom
i
c
net
work ma
nag
ement. There
ontologie
s
a
r
e
use
d
to a
u
g
m
ent the fa
ct
s rep
r
e
s
ente
d
in in
dep
end
ent informatio
n and
data
m
odel
s ad
ding
the
adeq
uate se
mantics
that enabl
es
the
mappin
g
of t
heir
co
mmon
voca
bula
r
y i
n
to the
network
element
s
h
e
terog
ene
ou
s manag
eme
n
t interfaces
an
d data
mo
del
s. T
he
refe
re
nce
[11] p
r
e
s
ents
a virtuali
z
ed
sol
u
tion
by mea
n
s of
virtual
ma
chi
ne live
migration a
p
p
r
oa
ch to
e
nha
n
c
e
availability, reso
urce m
a
n
ageme
n
t, po
wer man
age
ment, and
fa
ult-toleran
c
e.
At pre
s
ent,
the
method
h
a
s been widely use
d
in
the rese
arch
o
n
system reli
abil
i
ty and availa
bility, and it has
become hot
spot in resea
r
ch an
d appli
c
ation with
mu
ltidisci
plina
r
y cro
s
s ch
aract
e
risti
cs. Ba
se
d
on a
u
tono
mic co
mputin
g p
r
inci
ple, h
a
ving ove
r
all
de
pend
ability a
s
the
g
oal, from the
lo
cal f
i
ne-
tuning of th
e layers in
depe
ndability
eleme
n
ts, t
h
is p
ape
r a
dopts th
e m
e
thod of li
n
ear
program
m
ing to implement the dependability optim
iz
ation in the layers
of IOT, building an
autonomi
c
op
timization mo
del of IOT system.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 1, March 2
016 : 326 – 3
3
4
328
2. Modeling and analy
s
is
2.1. Dependabilit
y
Index Extraction
In each l
e
vel, dependabilit
y
factor
i
s
div
e
rse. Such
as inform
ation i
n
perception l
a
yer
will
experie
nce the pro
c
e
s
s flow in
cludin
g
perceptio
n,
acqui
sition, ga
thering, fusi
o
n
, transmi
ssi
on,
stora
ge, mini
ng, deci
s
io
n-makin
g
an
d control, a
s
a result, pe
rcept
ion criterio
n a
ffecting this la
yer
depe
ndability
come from
depe
ndability
of sen
s
ing
node,
resource rest
riction
of sen
s
ing
and
gatheri
ng poi
nts,
depen
dability of information
collection, priv
acy
of inform
ation transmi
s
sion, to
avoid the po
ssi
ble de
pen
dability probl
em, such
as node camou
f
lage, leakag
e of signal
s
and
interferen
ce, damag
e
to sensi
ng software and ha
rd
ware, un
auth
o
rized
use, perceptio
n d
a
ta
theft. As
the c
o
re data t
r
ans
miss
ion layer of
IO
T, Net
w
ork
criterion of
network
layer need
con
s
id
er de
p
enda
bility and depen
dabil
i
ty of Ne
twork, dep
end
a
b
ility of data
and priva
cy,
reliability of router p
r
oto
c
o
l
, to withstan
d tr
an
smi
ssi
o
n
band
width
occ
upie
d
,
rapid spread of
dependability
threat, message the
ft, message distorting,
prot
ocol damage, high energy
con
s
um
ption,
etc. Acco
rding to the diffe
rence of application domain a
nd mana
ge
ment
mech
ani
sm,
appli
c
ation
criterion
of a
p
p
lication
la
yer
are
re
stri
cted
by Se
rvice
i
ndu
stry, a
c
ce
ss
control, information sto
r
ag
e and man
a
g
e
ment mo
d
e
, which inclu
d
e servi
c
e type, service obj
ect,
hetero
gen
eo
us n
e
two
r
k
authenti
c
atio
n, attack of
virus/ha
cke
r
/malwa
re, ille
gal usage
of 3G
terminal
s, etc..
Extraction
an
d ab
stra
ct th
e critical
ele
m
ents
of the
system
de
p
enda
bility fro
m
the
multiple de
pe
ndability ele
m
ents a
nd m
appin
g
them
i
n
to the de
pe
ndability inde
x of system
is the
first step
of the work.
This p
ape
r
sele
cts the
system dep
e
ndability ind
e
x and the
key
depe
ndability
elements a
s
follows.
(1) Su
rvivabil
i
ty describe
s
the relia
bility of t
he syste
m
in the
ca
se
of
rando
m failure
of
comp
one
nts.
Survivability
depe
nd
s n
o
t
only on
the t
opolo
g
ical
structure of
the system,
but a
l
so
on the fa
ult p
r
oba
bility of system compo
nents,
exte
rn
al fault an
d repair strategi
es. It is m
a
in
ly
affected
by t
he topologi
cal connectivity, fault
tolerance degree,
network
equilibrium
degree,
coh
e
si
on, en
d-to-end
relia
bility,
K
terminal reliability, all termi
nal
reliability, rout
e coverage and
busi
n
e
ss p
e
rf
orma
nce.
(2)
De
pen
d
ability inclu
d
e
s p
h
ysi
c
al
depe
nda
bility, data depend
ability, netwo
rk
depe
ndability
and
ap
plication d
epe
nda
bi
lity, which
in
cl
ude
s the
sy
stem a
b
ility to
anti sea
r
ch, a
n
ti
interception,
anti dire
ction
a
l analysi
s
, a
n
ti cheatin
g a
nd anti extern
al invasio
n
.
(3)
Completion refers
to the s
y
s
t
em ability
to acco
mp
lish the
syste
m
se
rvice
req
uest b
y
norm
a
l op
era
t
ion or deg
ra
ded
se
rvice
a
t
any mome
n
t
of a spe
c
ified
task, whe
n
the ta
sk st
arts
and the
avail
ability is ce
rt
ain. It is mai
n
ly reflec
te
d
in the thro
ug
hput, pa
cket loss rate, d
e
l
a
y,
band
width util
ization, re
sp
o
n
se time an
d resou
r
ce utilization.
(4) Availabilit
y is the ability of
the system to
maintain workable st
ate at any ti
me within
the pre
s
cribe
d
perio
d an
d unde
r sp
e
c
ified conditi
ons. The m
a
in ba
sic
pa
rameters in
cl
ude
informatio
n c
o
llectio
n rate,
erro
r rate, bl
oc
ki
ng rate, tran
smi
ssi
on delay, throug
hput, the number
of con
c
urre
nt use
r
s,
softwa
r
e fault tolera
nce, etc.
2.2. Optimization Model
The d
epe
nd
ability factor of IOT h
a
s
obviou
s
multilayer
and m
u
ltidi
m
ensi
onal
characteri
stics, which make th
e abilit
y to self
configuratio
n and self adj
ustment of I
O
T
depe
ndability
is limited. Therefore, in
this
ca
se,
depe
ndability
regul
ation i
n
one
step
is
impossibl
e. Base
d on t
he
depe
ndability
index of IOT
system, a
u
to
nomic compu
t
ing is fu
se
d i
n
to
singl
e-layer collaborative fine-
tuning process of users th
at will direct
ly decide the
comp
re
hen
si
ve depe
nda
bility of system, to
i
m
pleme
n
t the si
ngle
-
la
yer configu
r
ation
optimizatio
n. On this ba
sis, multi-layer syst
em de
pe
ndability adju
s
tment is imp
l
emented. From
microsco
pic to glo
bal
pe
rspe
ctive, the
auton
om
i
c
config
uratio
n
and
a
d
just
ment, from
single
layer to mul
t
i layer, can
achieve th
e self
-ren
ew and optimi
z
ation of th
e whol
e system
depe
ndability
configu
r
ation
.
This paper researches the dep
endabilit
y insurance
method fr
om
the perspective of the
optimal
re
sou
r
ce
s configu
r
ation. By the
key
point
or the
wea
k
lin
k fo
r
the
syst
em
comp
ositi
on
stru
cture, we
use li
nea
r p
r
o
g
rammi
ng
an
d multid
ime
n
s
ion
a
l un
co
nstrained
optimi
z
ation
pri
n
cipl
e
to co
nsid
er si
ngle-l
a
yer
an
d glob
al d
e
p
enda
bility pro
b
lem of th
e system; then
we
ca
rry o
n
t
h
e
optimizatio
n of system confi
guration
and a
c
hieve
the final pur
po
se to pro
t
ect the syst
em
depe
ndability
through
con
f
iguring the d
epen
dability fa
ctors of intra layer and i
n
ter layer. T
he
system o
p
timization mo
del
is sh
own in Figure 1.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
An Autonom
ic Optim
i
zatio
n
Model of M
u
lti-La
ye
red
Dep
end
ability for… (Zh
eng
Ruijua
n
)
329
Figure 1. System Optimiza
tion Model
3
.
Optimal Configura
t
ion
3.1. Mapping bet
w
een Dependabilit
y
I
ndex and
Dependabilit
y
Element
Among d
epe
ndability ele
m
ent of p
e
rception laye
r(
E
P
),information coll
ection rate(
Cr
),
coh
e
siv
e
ne
ss
(
Ch
), termi
n
al pair
reliability(
Tr
)
and resource ut
ilization(
Ru
)
will be the
key
element
s affecting the surv
ivability of perception
(
SV
p
), safety of
pe
rce
p
tion
(
SF
p
), co
mpletio
n
of
perceptio
n(
CP
p
) and
ava
ilability of perception
(
AB
p
) of this l
a
yer. So, the survivability o
f
perceptio
n
SV
p
is restri
cted by terminal
pair reli
abilit
y
Tr
and information collection rate
Cr
, a
nd
the cal
c
ulatio
n formula i
s
a
s
formul
a (1
).
1
1
11
12
1
1
,
pi
p
i
i
ii
T
SE
r
VC
r
(1)
Safety of perceptio
n
SF
p
is restri
cted by terminal pai
r reliability
Tr
, reso
urce utili
zation
Ru
an
d co
he
sivene
ss
Ch
,
and the calcu
l
ation formul
a
is as form
ula
(2).
2
2
21
22
23
2
,1
ip
i
i
ii
p
SF
T
r
ER
u
C
h
(2)
Compl
e
tion o
f
perce
ption
CP
p
is re
stri
cted by info
rmation collection rate
Cr
a
nd
resource utilization
Ru
, an
d the cal
c
ulati
on formul
a is
as form
ula (3
).
3
3
31
32
3
1
,
ip
i
i
pi
i
CE
PC
r
R
u
(3)
Availability of perception
AB
p
is rest
ricted
by informatio
n coll
ection
rate
Cr
an
d
coh
e
siv
e
ne
ss
Ch
, and the
cal
c
ulatio
n formula i
s
as fo
rmula (4).
44
4
1
4
2
4
1
,
ip
i
i
pi
i
AE
BC
r
C
h
(4)
3.2. Local Optimization
Defini
tion 1.
Optimization c
o
s
t
(
OC
).
On the
prem
ise
of given
the initial val
ues to
depe
ndability
factors, the local optimi
z
ation in
lay
e
rs, value
s
of depen
dab
ility factors for
depe
ndability
maximizatio
n
will ch
ang
e, and the deviation of this ch
ang
e is call
ed
a
s
the
co
st
optimizat
ion. For e
x
ample, the depend
abil
i
ty factor set in busin
ess layer is
,,
{}
,
B
E
Ft
Ai
Nc
Ir
, if the initial value
s
of d
epe
ndability facto
r
s
se
parately
are
00
0
0
,,
,
F
tA
i
N
c
I
r
,
and the
optim
al co
nfiguration value
s
a
r
e
,,
,
O
C
OC
OC
OC
F
tA
i
N
c
I
r
, then the op
timization
co
st of this
local o
p
timiza
tion can b
e
e
x
presse
d as:
00
0
0
(|
|,
|
|
,
|
|
,
|
|
)
OC
OC
OC
OC
OC
Ft
Ft
A
i
Ai
N
c
N
c
I
r
I
r
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 1, March 2
016 : 326 – 3
3
4
330
Defini
tion 2.
Local
Optim
i
zation O
b
je
ctive (
LO
). Aiming at the goal to maxi
mize the
depe
ndability
in a layer, th
e maximum v
a
lue of t
he
differen
c
e b
e
tween o
p
timal configuration a
n
d
optimizatio
n
co
st is the l
o
cal
opt
imization o
b
je
ctive. To b
e
clea
r,
duri
ng th
e a
d
justme
nt in
a
layer, ‘imple
m
enting the
optimal confi
guratio
n with
the minimu
m co
st’ is th
e ideal g
oal
of this
adju
s
tment, while, b
e
cau
s
e the optimal
config
uratio
n
and o
p
timizat
i
on cost a
r
e b
o
th co
nst
r
ain
e
d
by the valu
e
of dep
end
abil
i
ty factors, th
e maxi
mum
d
i
fference b
e
twee
n the
two
is
ch
osen
as the
optimizatio
n obje
c
tive in the pra
c
tical a
p
p
licatio
n.
Defini
tion 3.
Varia
b
le el
ement
(0
1
)
.In addition to th
e
value
of a
few
key
element
s,
the remaini
ng va
riable
s
value
s
dete
r
minin
g
depen
dabilit
y index are called a
s
varia
b
le
element
s
(0
1
)
.The value of th
e depen
dabil
i
ty index decided by the comp
re
hen
si
ve
value
of key element
s
(0
1
)
and variable
elem
ents
is the
ide
a
l dep
end
abil
i
ty value 1,
that is
1
.
Defini
tion 4.
Quota
s
comp
ensation.
If
when a key
ele
m
ent
is pa
rticularly o
u
tstan
d
ing, i
t
can
effective
l
y comp
en
sa
te som
e
a
s
pect
s
of
the
system
de
pend
ability, then the
qu
o
t
as
comp
en
satio
n
of thi
s
key
element
ca
n
be u
s
e
d
to m
a
ke
up fo
r
promoting th
e
depe
ndability
of
sy
st
em.
S
u
c
h
as
3
Tr
C
h
or
21
Cr
R
u
.
Based o
n
ab
ove definition
s
, the pro
c
e
s
s of
local o
p
timization in p
e
rception lay
e
r ca
n be
cal
c
ulate
d
by the followin
g
formul
as.
(1)
Con
s
tru
c
ti
ng the optimi
z
ation o
b
je
ctive function.
11
2
2
3
3
4
4
()
,
,
,
)
(,
,
)
(
,
pp
p
p
i
Pp
p
p
p
p
SV
S
F
C
P
p
ij
p
j
k
p
k
l
p
l
ij
k
l
AB
ES
V
S
f
EE
E
E
FC
P
A
B
(5)
So, the optimization o
b
je
ctive in percepti
on layer is:
1
1
21
12
3
1
41
22
3
2
23
42
ma
x
(
)
(
)
()
)
(
(
)
pp
p
p
p
pp
p
p
PS
V
S
F
S
V
C
P
A
B
SF
CP
SF
AB
E
TC
Ch
fr
r
Ru
(6)
Q
u
ot
as
com
p
ens
a
ti
o
n
,,
,
,
.
,0
.T
,
Tr
Cr
Ru
Ch
Tr
Cr
R
u
Ch
Tr
Cr
R
u
S
Ch
Additionally,
12
3
4
,
1,
1,
1,
1,
(
0
,,
)
1
,
1
pp
p
p
SV
SF
CP
A
lB
ij
k
.
(7)
(2) Stand
ardi
zation
The optimi
z
at
ion functio
n
is normali
ze
d a
s
:
11
21
12
31
4
1
22
32
23
42
mi
n
(
)
()
(
)
()
pp
p
p
p
pp
p
p
SV
SF
SV
CP
A
B
SF
C
P
SF
AB
Tr
Cr
z
Ru
Ch
(7)
()
P
zf
E
(8)
4. Example Analy
s
is in
Percep
tion L
a
y
e
r
Maintainin
g the normalize
d
obje
c
tive functi
on u
n
cha
nged, the qu
otas compe
n
s
ation in
the optimal
co
ndition
s ca
n b
e
specifie
d a
s
)1
2)
3
Tr
Ch
Cr
R
u
((
, then th
e
optimizatio
n condition i
s
:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
An Autonom
ic Optim
i
zatio
n
Model of M
u
lti-La
ye
red
Dep
end
ability for… (Zh
eng
Ruijua
n
)
331
3
2
,,
,
,,
.T
.
,0
1
T
r
Cr
R
u
Ch
Tr
C
h
Cr
R
u
T
r
Cr
R
u
Ch
Tr
C
R
u
C
S
rh
In order to re
flect the gene
ral, the weigh
t
12
3
4
,,
,
,
,,
,
pp
p
p
SV
SF
CP
AB
i
j
k
l
a
nd
the value
of
key factors
(0
1
)
in the o
b
je
ctive functio
n
a
r
e
random
a
ssi
g
ned. To
be
sure,
the co
mpreh
ensive val
ue
of key facto
r
s
(0
1
)
is bigg
er th
an that of va
riable
elem
en
ts
(0
1
)
, s
o
, in this
example
0.
5
. Based o
n
ab
ove
con
d
ition
s
,
the optimal
sol
u
tion of
the obj
ective
function
z
in
an
y con
d
ition
s
i
s
cal
c
ulate
d
.
And the
calculation
re
sult
s a
r
e
sho
w
ed
in Table 1.
Table 1. Cal
c
ulation re
sult
s of the exam
ple
p
SV
p
SF
p
CP
p
AB
11
12
21
22
23
31
0.532
0.167
0.221
0.08
0.674
0.326
0.846
0.1
0.054
0.579
0.315
0.256
0.12
0.309
0.258
0.742
0.289
0.543
0.168
0.426
0.567
0.023
0.125
0.285
0.563
0.437
0.659
0.232
0.109
0.349
0.257
0.367
0.313
0.063
0.736
0.264
0.159
0.101
0.74
0.601
0.359
0.063
0.213
0.365
0.144
0.856
0.265
0.507
0.228
0.389
0.467
0.174
0.222
0.137
0.289
0.711
0.164
0.058
0.778
0.689
0.356
0.251
0.066
0.327
0.678
0.322
0.013
0.165
0.822
0.156
0.369
0.222
0.101
0.308
0.103
0.897
0.749
0.111
0.14
0.365
0.746
0.116
0.105
0.033
0.819
0.181
0.722
0.022
0.256
0.356
0.458
0.259
0.036
0.247
0.568
0.432
0.322
0.368
0.31
0.643
0.583
0.211
0.046
0.16
0.458
0.542
0.268
0.253
0.479
0.364
In Table 1, whe
n
the value of all wei
ghts an
d
are arbitrarily set in the ran
ge of
effective value, the fou
r
key ele
m
en
ts in p
e
rcept
ion layer ca
n se
ek the
optimal
solut
i
on
according
to
the n
eed
s
of obje
c
tive
depe
ndability
.
Mean
while,
be
cau
s
e
th
e sele
cted
q
uota
compensation item i
s
special,
the value of dependability element
R
u
i
s
al
way
s
0,
whi
c
h
do
es
not affect the cal
c
ulatio
n
of local opti
m
ization
o
b
je
ctive on the basi
s
of the optimal sol
u
tion.
Based
on
the
re
sults,
and
according
to
the con
c
eptio
n of O
p
timiza
tion Cost in
d
e
finition 1
an
d
Local Optimi
zation
Obje
ctive in defini
t
ion 2, we
d
e
rive the
cal
c
ulatio
n
form
ula of the l
o
cal
optimizatio
n obje
c
tive
is
as
follows
:
00
O
C
OC
0
0
0
0
OC
0
0
O
C
,
,,
(|
2
|
,
)
,
,
,
,
;
|2
|
,
,
.
OC
OC
ii
i
i
i
ep
j
j
k
j
j
j
k
k
k
jk
ii
i
i
i
i
ep
e
p
Ep
ep
e
p
L
o
ep
ep
e
p
ep
Ep
e
p
ep
ep
Ep
e
p
ep
ep
ep
e
p
Ep
ep
ep
;
sa
t
i
s
f
i
e
s
sa
t
i
sf
i
e
s
sa
t
i
s
f
i
e
s
sa
t
i
sf
i
e
s
(9)
32
41
42
Tr
Cr
R
u
Ch
f
Quotas
compensation
0.421
0.586
0.414
0.05 0.5
0
0.25
0.8
0.2097
3
Tr
C
h
21
Cr
R
u
0.574
0.268
0.732
0
0.6178
0
0.3089
0.612
0.3103
0.651
0.458
0.542
0.005
0.5
0
0.25
0.755
0.2519
0.399
0.364
0.636
0
0.5342
0
0.2671
0.71
0.2322
0.611
0.458
0.542
0.147
0.5
0
0.25
0.897
0.3418
0.311
0.514
0.486
0.194
0.5
0
0.25
0.944
0.3599
0.844
0.898
0.102
0.245
0.5
0
0.25
0.995
0.3291
0.635
0.566
0.434
0
0.5521
0
0.276
0.689
0.3448
0.644
0.823
0.177
0.106
0.5
0
0.25
0.856
0.1823
0.357
0.589
0.411
0.018
0.5
0
0.25
0.768
0.2349
0.636
0.521
0.479
0.062
0.5
0
0.25
0.812
0.2725
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 1, March 2
016 : 326 – 3
3
4
332
Her
e
,
12
1
[
,
,
.
..,
,
,
.
..,
]
,
[
1
,
.
..
]
,
[
1
,
...
]
,
[
1
,
...
]
mm
n
E
p
ep
ep
ep
ep
ep
i
n
j
m
k
m
n
.
The i
n
itial v
a
lue
0
([
1
,
.
.
.
]
)
i
ep
i
n
of opti
m
ization
obj
ective i
s
a
s
sign
ed
ran
d
o
mly.
Becau
s
e
the
optimal valu
e of
R
u
is al
way
s
zero, we choo
se th
e local optimal
objective only
con
s
id
erin
g the value ch
a
nge
s of
Tr
,
Cr
and
Ch
.Values of the optimal obj
ectives a
r
e showed
in Table 2.
Acco
rdi
ngly,
discrete
poi
nt map
s
of th
e
rand
om
initial
value
s
a
r
e
shown in
Figu
re 2. Th
e
initial values
of discrete po
ints are
widel
y distributed.
Figure 2. Discrete p
o
int of
0
Tr
,
0
Cr
and
0
Ch
Table 2. Co
rresp
ondi
ng lo
cal optimal o
b
jective of
0
([
1
,
.
.
.
]
)
i
ep
i
n
0
Tr
0
Cr
0
Ch
f
0
Tr
Cr
Ch
Lo
0.0346
0.6608
0.1265
0.282924
0
0.6178
0.3089
0.7359
0.2646
0.4218
0.3149
0.405561
0
0.6178
0.3089
0.9893
0.1285
0.5216
0.2516
0.388233
0
0.6178
0.3089
0.9017
0.1483
0.7683
0.3516
0.429906
0
0.6178
0.3089
0.8818
0.0869
0.2549
0.5726
0.296562
0.147
0.5
0.25
0.4144
0.1586
0.0346
0.3186
0.158572
0
0.6178
0.3089
0.4924
0.6526
0.2156
0.0542
0.433038
0
0.5342
0.2671
0.9224
0.0864
0.4682
0.3107
0.392218
0
0.6178
0.3089
0.8617
Before optimi
z
ation, the value dist
ributi
on of
0
Tr
,
0
Cr
and
0
Ch
is with representative
signifi
can
c
e,
and the valu
e
s
of
Tr
,
Cr
and
Ch
after optimi
z
atio
n sho
u
ld
con
v
erge the di
screte
situation of the initial value to a
certain extent. The compa
r
ing bet
wee
n
0
Tr
,
0
Cr
,
0
Ch
and
Tr
,
Cr
,
Ch
is sho
w
ed in
Figure 3. Here, the values
of
Tr
,
Cr
and
Ch
co
nverge the initia
l value.
In 11
ca
se
s
of Table
1, t
o
ea
ch
set o
f
assign
ed v
a
lue
s
0
([
1
,
.
.
.
]
)
i
ep
i
n
, the op
timal
obje
c
tive correspon
ding t
o
the optimal
solu
tion can
be calculate
d
. For exam
ple, when
0
Tr
=0.26
46,
0
Cr
=0.
4218 a
nd
0
Ch
=0
.3149, the o
p
timal obje
c
tive
Lo
=0.98
93, then the opti
m
al
solut
i
o
n
Tr
=0,
Cr
=0
.6178,
Ch
=0.308
9. So, the
opt
imization
effe
ct in
the
co
nd
itions
of a
s
sig
ned
initial value is shown in Figure 4.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
An Autonom
ic Optim
i
zatio
n
Model of M
u
lti-La
ye
red
Dep
end
ability for… (Zh
eng
Ruijua
n
)
333
Figure 3. Co
mpari
ng bet
ween
0
Tr
,
0
Cr
,
0
Ch
and
Tr
,
Cr
,
Ch
(a)
Comp
ari
n
g of optimizat
ion effect bet
wee
n
0
Tr
,
0
Cr
,
f
0
and
Tr
,
Cr
,
Lo
(b)
Comp
ari
n
g of optimizat
ion effect bet
wee
n
0
Cr
,
0
Ch
,
f
0
and
Cr
,
Ch
,
Lo
(c) Com
p
a
r
in
g of optimizat
ion effect bet
wee
n
0
Tr
,
0
Ch
,
f
0
and
Tr
,
Ch
,
Lo
Figure 4. Co
mpari
ng bet
ween optimi
z
at
ion effect in p
e
rception lay
e
r
In Figure 4
(
a
)
, (b), (c) se
pa
rately expre
s
s
the com
p
a
r
i
ng of optimization effect b
e
twee
n
0
Tr
,
0
Cr
,
f
0
and
Tr
,
Cr
,
Lo
,
the compa
r
ing
of optimizatio
n effect
between
0
Cr
,
0
Ch
,
f
0 and
Cr
,
Ch
,
Lo
,the compa
r
i
ng of optimization effect b
e
twee
n
0
Tr
,
0
Ch
,
f
0
and
Tr
,
Ch
,
Lo
.
We
c
an
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 1, March 2
016 : 326 – 3
3
4
334
see that, wh
en
0
Tr
=0.1
483,
0
Cr
=0.7683,
0
Ch
=0.35
16, the initial
depen
dabilit
y is
f
0
=0.429
906,
and the
opti
m
ized
pe
rformance rea
c
h
e
s 0.8
818, th
at is to
say, the de
pen
dab
ility in percep
t
ion
layer is o
p
timized
by 1.08 times; wh
e
n
0
Tr
= 0.15
86,
0
Cr
= 0.0346,
0
Ch
= 0.3186, the ini
t
ial
depe
ndability
is
f
0
= 0.15
85
72, an
d the
o
p
timized
pe
rf
orma
nce i
s
0
.
4924,
whi
c
h
is in
crea
sed
by
2.1 times.
5. Conclusio
n
Based
on l
a
yered th
oug
h
t
in IOT, thi
s
pa
pe
r ha
s integrated
autonomi
c
computing
con
c
e
p
t to the auton
omic
optimizatio
n
pro
c
e
ss
of th
e syste
m
. Ha
ving perce
ption layer
as t
h
e
example
an
d u
s
ing
line
a
r
pro
g
ra
m
m
ing m
e
tho
d
to
re
sea
r
ch an
d
set
up a
n
a
u
ton
o
mic
optimizatio
n
model
orie
nted to intelli
g
ent IO
T syst
em.
Havin
g
the
chang
e of
depe
nda
bi
lity
element
s directly affectin
g the depen
dability of sy
stem as the
internal
cau
s
e, based o
n
the
fusion
re
sult
s of de
pend
a
b
ility element
s, the
com
p
l
e
x multi-sou
r
ce
dep
enda
b
ility param
eters
variable
s
, co
nfiguratio
n variabl
es of I
O
T and d
e
p
enda
bility environme
n
t variable
s
have
been
abstracte
d. Using the li
nea
r pro
g
rammin
g
met
hod to
derive the m
appin
g
relatio
n
shi
p
betwee
n
the de
pen
da
bility element
s a
nd th
e d
e
pend
ability in
pe
rception
la
yer, the
opti
m
ization
mo
d
e
l of
intelligent IO
T ha
s
bee
n
obtaine
d. At the
sam
e
ti
me of l
o
cal
config
uratio
n
optimi
z
ation
in
perceptio
n la
yer, the co
mplete
syste
m
co
nfi
guration coo
r
dinati
on ha
s b
e
e
n
pe
rform
e
d
to
prom
ote the
overall
syste
m
de
pend
abil
i
ty. The la
st
analysi
s
re
su
lts
of exampl
es sh
ow
that,
in
the con
d
ition
of rando
m assignm
ent to key
elem
e
n
ts, the ave
r
age
optimization ra
nge
of
prop
osed mo
del is bigg
er
than 39%. Percepti
on lay
e
r ha
s impro
v
ed the depe
ndability, and
the
optimizatio
n effect is sig
n
i
f
icant. In the future
work, we will focu
s o
n
the global coordi
nation a
n
d
optimizatio
n.
Ackn
o
w
l
e
dg
ements
This work h
a
s
b
een
supp
orted
by the
Proje
c
t of
Na
tural S
c
ien
c
e
Fou
ndation
of Chi
n
a
(No.
U1
2046
14, 613
702
2
1
,U14
046
11),
in pa
rt by
Prog
ram fo
r In
novative Re
search T
eam
(in
Scien
c
e and
Tech
nolo
g
y) in
University of
He
nan
Pro
v
ince
(No. 1
4
I
RTSTHN02
1
)
, in
part
by K
e
y
Proje
c
t of Science and Te
chnolo
g
y Dep
a
rtm
ent of He
nan Provin
ce
(No.1
121
022
1018
7).
Referen
ces
[1]
Li F
,
Ya
ng J,
W
u
J, An C,
Jian
g N.
Re
se
arch on autom
atic
co
nfigur
ati
on of
Intern
et.
Jour
nal
o
f
Software
. 2014
; 25(1): 118-1
3
4
.
[2]
Gonen
B, Yuks
el M.
N
e
tw
ork config
uratio
n a
nd mana
ge
me
nt
via
tw
o-phas
e o
n
li
ne
opti
m
i
z
a
t
i
on.
Pr
oc.
of Global T
e
lec
o
mmunic
a
tio
n
s
Confere
n
ce (
G
LOBECOM 2011). 20
11: 1-6
.
[3]
Peng M, Lia
n
g
D, W
e
i Y,
and L
i
J. Self-conf
igur
ation
and self-o
pti
m
iz
atio
n in L
T
E-advanced
hetero
g
e
neo
us
net
w
o
rks.
Co
mmu
n
icati
ons Maga
z
i
ne
. 2
0
1
3
; 51(5): 36-4
5
.
[4]
Li Y, F
e
ng Z
,
Xu
D, Z
h
ang
Q,
T
i
an H.
A
u
tomat
ed opti
m
al
co
nfig
urin
g
of femt
ocel
l b
a
se stati
ons'
para
m
eters in
enterpr
ise fe
mtoce
ll n
e
tw
ork
. Proc. of Globa
l T
e
leco
mmunicati
ons
Confer
enc
e
(GLOBECOM
201
1). 201
1: 1-5.
[5]
Lopes MC, A
Dias B.
Auto
mated n
e
tw
ork services c
onfig
uratio
n
man
a
g
e
ment
. Proc.
of IFIP/IEEE
Internatio
na
l Symp
osi
u
m on Integrate
d
Ne
t
w
ork Mana
geme
n
t-W
o
rkshops.
200
9: 140-
143.
[6]
Bjorklu
nd M. YANG-A data mode
lin
g lan
g
u
a
ge for
the net
w
o
rk config
ura
t
i
on protoc
ol (NET
CONF
),
RFC60
20. 20
1
0
.
[7]
Lop
es M, Costa A, Dias B.
Impr
ovin
g netw
o
rk services re
silie
nc
e thro
ug
h auto
m
atic s
e
rvice n
o
d
e
config
uratio
n g
ener
ation
. Pr
oc
eed
ings
of Net
w
o
r
k Oper
ation
s
and M
a
n
age
ment S
y
mpos
i
u
m (NOMS).
201
2: 919-
925.
[8]
Elbada
w
i
K, Yu J.
I
m
prov
in
g n
e
tw
ork ser
v
ices c
onfig
ur
ation
ma
nag
e
m
e
n
t
. Proce
e
d
i
ngs
of
20t
h
Internatio
na
l C
onfere
n
ce o
n
Comp
uter
Com
m
unic
a
tions a
n
d
Net
w
orks. 20
11: 1-6.
[9]
Eissa A. F
eature Sel
e
ctio
n
Algorithms f
o
r W
i
reless S
ensor N
e
t
w
ork
s
.
In
te
rn
a
t
i
o
nal
Jo
u
r
na
l
of
Information &
Network Security
. 2014; 3(5):1
-7.
[10]
Strassner J, A
gou
lmin
e N,
L
ehtih
et E. F
o
c
a
le: A
n
o
vel
a
u
tonom
ic n
e
t
w
orkin
g
arc
h
itec
ture.
ITSSA
Journ
a
l
. 20
07; 3(1): 64-7
9
.
[11]
Sang
eeta
S. A
Revi
e
w
on
Ef
ficient V
i
rtual
Ma
chi
ne Liv
e
Migratio
n:
C
h
a
llen
ges,
r
e
q
u
ir
ements an
d
techno
lo
g
y
of VM migratio
n i
n
clou
d.
Intern
ation
a
l Jour
na
l
of Cloud
C
o
mputin
g an
d Ser
v
ice Scie
nce
.
201
4; 3(6): 1-8
.
Evaluation Warning : The document was created with Spire.PDF for Python.