Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
6, N
o
. 3
,
Ju
n
e
201
6, p
p
. 1
031
~ 10
37
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
3.9
041
1
031
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Review of Software Fault-Tol
erance Methods for Reliability
Enhancement of Real-Time Software Systems
Anj
u
shi Verm
a, An
kur Gh
a
a
rt
an
, T
i
rth
a
nkar
G
aye
n
Com
puter
and S
y
s
t
em
s
S
c
i
e
nces
,
J
a
waharlal Nehr
u University
, Ind
i
a
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Sep 16, 2015
R
e
vi
sed M
a
r
7,
2
0
1
6
Accepted
Mar 20, 2016
Real
time s
y
stems are those sy
stems
which
must guarantee to response
correc
t
l
y
with
in
strict t
i
m
e
constraint or wi
thin
deadl
i
ne. Fa
ilur
e
s can ar
ise
from both functional errors
as well
as tim
ing bu
gs. Hence, it is
necessar
y
to
provide temporal correctness of programs
used in real time applications in
addition
to pro
v
iding function
a
l corr
ectness.
Although, th
ere are several
res
earch
es
con
c
erned with
a
c
h
i
eving f
a
ult
tol
e
ranc
e in
the
pres
ence
of
various function
a
l
and oper
a
tion
al
errors
but man
y
of
th
em did
not addr
ess
the prob
lem con
cerned
with
the
timing
bugs which is
an important issue
in
real
tim
e s
y
s
t
e
m
s
.
As
for rea
l
tim
e s
y
s
t
em
s
,
m
a
n
y
tim
es
it
becom
e
s
a
necessit
y
for
a
given servi
ce
to be de
liver
ed
within th
e spe
c
ifi
e
d tim
e
deadline. Ther
ef
ore, th
is paper
revi
ews
the
exi
s
ting approa
che
s
from
the
pers
pect
ive of
real
tim
e s
y
s
t
e
m
s
to anal
ys
e
the s
hortcom
ing
s
of thes
e
approach
es
to pr
es
ent a vers
ati
l
e
and cos
t
eff
ect
iv
e approa
ch in th
e pres
enc
e
of tim
ing bugs f
o
r providing f
a
u
lt tol
e
ran
ce
to enhance
the r
e
l
i
abilit
y
o
f
the
real
tim
e
s
o
ftwar
e
app
lic
at
ions
.
Keyword:
Fau
lt to
leran
c
e
Real tim
e
syste
m
s
Reliab
ilit
y
Soft
ware
Copyright ©
201
6 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
An
jushi
Ve
rm
a,
Com
puter and
Syste
m
s Sciences,
Jawa
harl
al
Ne
hr
u
Uni
v
er
si
t
y
, I
ndi
a.
1.
INTRODUCTION
A re
al
-t
im
e sy
st
em
i
s
one
t
h
at
m
u
st
pro
duce
a
co
rrect
resu
lt with
in a sp
ecified ti
me deadline
.
According to
Herm
ann Kopetz, “A real
time co
m
puter s
y
ste
m
is a comput
e
r
system where the correc
t
ness of
t
h
e sy
st
em
beh
a
vi
o
r
de
pe
nds
not
onl
y
o
n
t
h
e
l
ogi
cal
res
u
l
t
s
of t
h
e com
put
a
t
i
ons,
b
u
t
al
so
on t
h
e p
h
y
s
i
cal
t
i
m
e
whe
n
these
re
sults are produ
ced”
[1]. Some exa
m
ples of
real tim
e
s
y
ste
m
s are aircraft control,
ove
n
t
e
m
p
erat
ure
c
ont
rol
l
e
r a
n
d
ove
r-t
em
perat
u
re m
oni
t
o
r i
n
nuclear
powe
r station, et
c.Failu
res i
n
real ti
me
syste
m
s can
arise fro
m
bo
th fu
n
c
tion
a
l erro
rs as
we
ll as ti
m
i
n
g
bug
s.
Hen
c
e, it is
necessary to pro
v
i
d
e
te
m
p
o
r
al co
rrectn
e
ss
o
f
pro
g
ram
s
u
s
ed
in
real tim
e
ap
p
lication
s
i
n
ad
d
ition
to p
r
ov
id
ing
fun
c
tio
n
a
l
correctness
.
Although, the
r
e a
r
e se
vera
l rese
arches
conce
r
ned with
ac
hiev
in
g
fau
lt to
leran
ce in th
e presen
ce
of
vari
ou
s f
unc
t
i
onal
an
d o
p
e
r
at
i
onal
err
o
rs b
u
t
m
a
ny
of t
h
e
m
di
d n
o
t
ad
dr
ess t
h
e p
r
obl
e
m
concer
ne
d w
i
t
h
t
h
e
ti
min
g
bug
s wh
ich
is an
imp
o
rtan
t issue in
real ti
m
e
sy
ste
m
s. As for real ti
m
e
sys
t
e
m
s,
m
a
n
y
t
i
m
es it
becom
e
s a necessity for a gi
ven se
rvice to
be delive
r
ed
with
in
th
e sp
eci
fied
ti
m
e
d
ead
lin
e. Th
erefo
r
e, th
is
wo
rk
anal
y
s
es
t
h
e sh
ort
c
om
i
ng o
f
t
h
e e
x
i
s
t
i
ng a
p
pr
oac
h
es
with res
p
ect t
o
real tim
e sys
t
e
m
s and prese
n
ts a
v
e
rsatile, co
st effectiv
e appro
ach
in
th
e presen
ce of ti
m
i
n
g
b
u
g
s
fo
r pro
v
i
d
i
ng
fau
lt to
leran
ce to
real ti
me
so
ft
ware app
licatio
n
s
. “Soft
w
are
fau
lt to
leran
ce is t
h
e ab
ility o
f
co
m
p
u
t
er so
ft
ware t
o
con
tinu
e
its
n
o
rm
a
l
ope
rat
i
o
n des
p
i
t
e
t
h
e prese
n
c
e
of sy
st
em
or har
d
wa
re f
a
ul
t
s
.” Avi
z
i
e
ni
s [2]
i
n
19
7
7
d
e
fi
ne
d S
o
ft
wa
r
e
Faul
t
Tolera
nce as
“
t
he m
a
na
ge
me
nt
of
f
aul
t
s
ori
g
i
n
at
i
n
g
f
r
o
m
def
ect
s i
n
desi
gn
of
t
h
e s
o
f
t
w
are
.”
Reliab
ilit
y h
a
s always b
een
an i
m
p
o
r
tan
t
qu
ality
attr
ib
u
t
e o
f
so
ftware system
s
esp
ecially
fo
r m
i
ssio
n
and safety criti
cal soft
ware
syste
m
s
because in
suc
h
syste
m
s
seve
rity
of
c
onseque
nces
re
sulting from
failures
i
s
very
hi
g
h
.
Acco
r
d
i
n
g t
o
AN
SI, “
So
ftwa
r
e Relia
b
ility i
s
th
e p
r
o
bab
ility o
f
fa
ilu
re-free so
ftwa
r
e opera
tion
fo
r a
sp
ecified p
e
riod
o
f
time in
a
sp
ecified
en
viro
nmen
t.
” Soft
ware is
becom
i
ng m
o
re and m
o
re co
m
p
lex
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E
V
o
l
.
6,
No
. 3,
J
u
ne 2
0
1
6
:
10
3
1
– 10
37
1
032
with
th
e p
a
ssag
e
o
f
tim
e to
co
p
e
with
t
h
e emerg
i
ng
ap
p
licatio
n
requ
iremen
ts. M
o
re c
o
m
p
lex a soft
w
a
re is,
m
o
re d
i
fficu
lt it is to
assure its reliab
ility.
A co
m
p
lex
so
ft
ware
h
a
s a
larg
e
nu
m
b
er o
f
states (u
n
l
i
k
e th
e
har
d
ware
), so i
t
i
s
not
pract
i
cal
l
y
possi
bl
e t
o
com
p
l
e
t
e
l
y
t
e
st
a soft
ware
. Ir
respect
i
v
e
of t
h
e am
ount
of t
e
st
i
ng
o
n
e
do
es; it is u
s
ually d
i
fficu
l
t
to
assure th
at
th
e fin
a
l so
ft
ware produ
ct is fau
lt free. In
ord
e
r t
o
ach
iev
e
failu
re
free op
eratio
n
o
f
so
ftware,
o
n
e d
e
v
e
lop
s
some
mech
an
ism to
h
a
nd
le th
o
s
e fau
lts wh
ich
rem
a
in
in
th
e s
y
ste
m
ev
en
after its dev
e
lop
m
en
t. So
ft
ware fau
lt to
leran
ce is su
ch
a m
ech
an
is
m wh
ich
can
b
e
u
s
ed
to
d
eal with
th
e
rem
a
i
n
i
ng faul
t
s
aft
e
r devel
o
pm
ent
of sy
st
em
. Si
nce real
-t
im
e sy
st
em
s need t
o
be hi
g
h
l
y
rel
i
a
bl
e and as i
t
i
s
not
al
way
s
p
o
s
s
i
b
l
e
t
o
e
n
s
u
re
t
h
e
devel
opm
ent
o
f
hi
g
h
ly reliab
l
e syste
m
th
erefor
e
they are prim
e
candidates
for th
e in
clusion
o
f
fau
lt to
leran
ce tech
n
i
q
u
e
s.
2.
RELATED BA
CKGR
OU
ND
Seve
ral techni
que
s are c
u
rrently in use t
o
ac
hiev
e
fault to
leran
ce in so
ft
ware li
k
e
N-Version
Pro
g
r
am
m
i
ng, R
ecove
ry
B
l
ocks (R
cB
),
N-S
e
l
f
C
h
ecki
n
g P
r
o
g
ram
m
i
ng, N-C
opy
Pr
o
g
r
a
m
m
i
ng, R
e
t
r
y
B
l
ock
s
(R
t
B
), A
d
a
p
t
i
v
e N-
Versi
o
n S
y
st
em
s, R
e
juv
e
nat
i
o
n
,
F
u
zzy
Vot
i
n
g, et
c, y
e
t
t
h
ese t
echni
que
s m
a
y
not
be ve
ry
su
itab
l
e fo
r
real ti
me syste
m
s. Soft
ware fau
lt to
leran
ce tech
n
i
q
u
e
s are
m
o
stl
y
b
a
sed
o
n
t
r
ad
ition
a
l h
a
rdware
fau
lt to
leran
ce. Th
e go
al o
f
so
ft
ware fau
lt to
leran
ce t
ech
n
i
q
u
e
s is to
allo
w th
e syste
m
t
o
fu
n
c
tion
prop
erly in
the prese
n
ce
of softwa
re fa
ult
s
rem
a
ining i
n
the syst
em
afte
r co
m
p
letin
g
the d
e
v
e
lop
m
en
t o
f
software.
Un
lik
e
h
a
rdw
a
r
e
ju
st
red
und
an
cy is no
t enou
gh
t
o
deal w
ith
so
ftw
a
re fa
ults; s
o
m
e
f
o
rm
of
dive
rsity
is also
re
qui
red
al
on
g wi
t
h
red
u
n
d
a
n
cy
[
3
]
.
N-
Versi
on
Pr
o
g
ram
m
i
ng (N
VP) i
s
o
n
e o
f
t
h
e so
ft
wa
re f
a
ul
t
t
o
l
e
rance
t
echni
q
u
es
bas
e
d o
n
desi
g
n
d
i
v
e
rsity. El
men
dorf in
1
972
sugg
ested
the co
n
c
ep
t o
f
NVP and
later in
1
977
–19
78 Av
izien
i
s and Ch
en
devel
ope
d
i
t
[
4
]
.
T
o
ac
hi
eve
fa
ul
t
t
o
l
e
ra
nc
e, i
n
N
V
P
t
echni
que
a
deci
s
i
on m
echani
s
m
(DM
)
an
d
a
f
o
r
w
ar
d
reco
very
m
echani
s
m
i
s
used.
At
l
east
t
w
o
vari
a
n
t
s
o
f
a
pr
o
g
ram
whi
c
h are
i
n
depe
n
d
ent
l
y
desi
gne
d a
n
d
fu
nct
i
o
nal
l
y
equi
val
e
nt
are
de
vel
o
ped
fr
om
the com
m
on
specifications.
All these variant
s
called versi
o
ns are
ru
n i
n
paral
l
e
l
an
d
resul
t
s
p
r
od
uce
d
by
eac
h
versi
o
n
i
s
pa
ssed t
o
t
h
e
D
M
. Deci
si
o
n
M
echani
s
m
sel
ect
s t
h
e
b
e
st resu
lt after ex
am
in
in
g
all th
e resu
lts. No
wad
a
ys v
a
rious alternative
decision
m
ech
anism
s
are ava
ilable
fo
r
use
wi
t
h
N
V
P.
I
n
1
9
7
4
Ho
rni
n
g, et
al
.
i
n
t
r
od
uce
d
R
e
cove
ry
B
l
oc
ks
(R
cB
)
[
5
]
.
T
h
e basi
c
R
c
B
s
c
hem
e
com
p
rises of
an e
x
ecuti
ve,
an acce
pt
ance
test, prim
ary
and alternate t
r
y bl
ocks
(variants). Base
d
on t
h
e
acceptance
tests (AT) re
sult
RcB selects a
varia
n
t res
u
lt
a
n
d pa
ss it to the output
du
ring program
exe
c
ution.
By u
s
ing
th
e prim
ary altern
at
e (o
r t
r
y b
l
o
c
k) th
e RcB tech
niq
u
e
will in
itially try to
en
su
re th
e
AT
(e.g
.,
a test
is passe
d
base
d
on the
acceptability of a
re
sult of a
n
a
lternate). If the
pri
m
ary al
gorithm
’
s result di
d
not
pass
th
e AT, th
en
n-1
altern
ativ
es
will b
e
tried
(atte
m
p
ted
)
u
n
til an
altern
ate’s
resu
lts p
a
ss th
e AT. An
error o
ccurs
whe
n
a
n
y
of
t
h
e al
t
e
r
n
at
es c
a
n
not
pass
t
h
e AT
[
6
]
.
La
pr
i
e
, et
al
de
vel
o
ped
N
-
Sel
f
-C
h
ecki
n
g P
r
og
ra
m
m
i
ng
(NSC
P) w
h
i
c
h
i
s
a desi
gn di
v
e
rse t
echni
q
u
e.
A sel
f-c
hecki
ng
pr
o
g
ram
uses pr
og
ram
redun
da
ncy
t
o
che
c
k i
t
s
ow
n
be
havi
or
du
ri
n
g
e
x
ec
ut
i
o
n
.
T
h
e
NSC
P
har
d
ware
arc
h
i
t
ect
ure com
p
ri
ses o
f
fo
ur
co
m
ponent
s
gr
o
u
p
ed
i
n
two pairs in hot standby redunda
ncy,
where
one softwa
re varia
n
t is supp
orte
d by each hardware com
p
one
n
t
.
NSCP s
o
ftware contains a
com
p
arison algorithm
a
nd t
w
o
va
ri
ant
s
o
r
an
AT
an
d
one
va
ri
ant
o
n
eac
h
h
a
rdw
a
r
e
p
a
ir
[7
].
Data d
i
versity as a co
m
p
lemen
t
ary so
ft
ware fau
lt to
leran
ce strateg
y
to
d
e
si
g
n
d
i
v
e
rsity was
pr
o
pose
d
by
Am
m
a
nn a
nd
Kni
ght
,
1
9
87
[8]
.
Dat
a
di
ve
rsi
t
y
based
t
echni
que
s i
m
pl
em
ent
s
di
ve
rsi
t
y
at
t
h
e
in
pu
t d
a
ta. To
ach
iev
e
d
a
ta d
i
v
e
rsity ev
ery data d
i
v
e
rs
i
t
y
ba
sed t
ech
ni
q
u
e uses a dat
a
re
-
e
xp
ressi
on al
g
o
ri
t
h
m
(DR
A
).
DR
A
pr
o
duces
di
ver
s
e i
n
put
dat
a
s
e
t
s
t
h
at
are l
o
g
i
cal
l
y
equi
val
e
nt
. T
h
e
per
f
o
r
m
ance o
f
dat
a
di
ve
rsi
t
y
i
s
hi
g
h
l
y
de
p
e
nde
nt
on t
h
e
dat
a
re
-ex
p
r
essi
on
al
g
o
ri
t
h
m
.
The p
r
o
b
l
em
associ
at
ed wi
t
h
dat
a
di
ve
rse
approaches
is that all applications ca
n
not em
ploy data diversity because
it
is not possi
ble to find a
n
e
f
fective
Dat
a
re
-e
xp
res
s
i
o
n
al
g
o
ri
t
h
m
f
o
r
eve
r
y
a
p
pl
i
cat
i
on.
Am
m
a
nn a
n
d
Kni
ght
de
vel
o
pe
d
R
e
t
r
y
B
l
oc
ks
(R
t
B
)
whi
c
h i
s
one o
f
t
h
e dat
a
di
ve
rse so
ft
ware
fa
ul
t
t
o
l
e
rance t
echni
que
s. AT
and
back
wa
rd
reco
very
t
o
ac
hi
eve
faul
t
t
o
l
e
ra
nce
i
s
used
by
R
t
B
t
echni
q
u
e.
L
a
t
e
r Am
m
a
nn and
K
n
i
g
ht
al
so de
vel
o
pe
d
N
-
C
o
py
Pr
o
g
ra
m
m
i
ng
(NC
P
) as
a
dat
a
di
v
e
rsi
t
y
t
echni
que
. T
h
e
N
C
P t
ech
ni
q
u
e
uses a
forwa
r
d rec
ove
ry a
n
d
a decision m
echanism
to accom
p
lish fault tolera
nce. At least two copies
of
a program
and at lea
s
t one data re
-expressi
on algorithm
i
s
use
d
by
NC
P. Sy
st
em
i
npu
t
s
are p
a
ssed
t
o
a
DR
A i
n
or
der t
o
gene
rat
e
l
ogi
cal
l
y
eq
ui
val
e
nt
i
n
p
u
t
da
t
a
set
s
.
Using
in
pu
t the re-expressed
d
a
ta th
e co
p
i
es ex
ecu
te in
pa
rallel. A decision m
echan
is
m
works to
ex
amin
e th
e
resu
lts
o
f
t
h
e co
p
y
ex
ecu
tio
n
s
and
if th
ere exists a b
e
st
resu
l
t
, it selects it. NCP is co
n
s
i
d
ered
as a
d
a
ta
d
i
v
e
rse
c
o
mp
l
e
me
n
t
o
f
N
V
P
.
The new Soft
ware Fault
T
o
lera
nce
techniques
are Fu
zzy Vo
ting
,
B
y
zan
tin
e Fau
lt To
leran
ce,
Ada
p
t
i
v
e
N-
V
e
rsi
o
n Sy
st
em
s and
G
r
ap
h R
e
duct
i
o
n.
Ka
no
un
, K
., et al. [
9
]
conside
r
ed
m
odified classical N-
Versi
on sy
st
e
m
s by
i
n
cor
p
orat
i
n
g i
n
eac
h ve
rsi
o
n a
n
i
ndi
vi
d
u
al
wei
ght
fact
or
. It
con
s
i
d
er
s an a
d
apt
i
v
e
ap
pro
ach
to
m
o
d
e
l an
d
man
a
g
e
d
i
fferent q
u
a
lity lev
e
ls o
f
th
e v
e
rsio
n
s
. Th
is weig
h
t
factor is th
en
i
n
co
rp
orat
e
d
i
n
t
h
e
vot
i
ng
p
r
oce
d
ure, i
.
e. f
o
r t
h
e de
vi
at
i
o
n be
ha
vi
o
r
o
f
t
h
e i
ndi
vi
d
u
al
versi
on t
h
e
vo
t
i
ng i
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Review of
Soft
w
are F
a
ult-Tol
e
rance Methods for
Reliab
ility Enhance
m
e
n
t of Re
al-Ti
m
e .... (A
njus
hi Ver
m
a)
1
033
base
d o
n
a we
i
ght
ed c
o
unt
i
n
g o
f
t
h
e n
u
m
b
er of m
oni
t
o
re
d eve
n
t
s
. I
n
o
r
der t
o
deal
wi
t
h
t
r
ansi
e
n
t
so
f
t
ware
failures ca
use
d
by
soft
ware a
g
in
g so
ftwa
re
reju
venatio
n i
s
a no
vel
ap
pr
o
ach,
whi
c
h ca
n be c
o
n
s
i
d
e
r
e
d
as a
pr
oact
i
v
e a
n
d
pre
v
e
n
t
i
v
e s
o
l
u
t
i
o
n
t
o
c
o
unt
eract
t
h
e a
g
i
n
g
phe
n
o
m
e
non
. R
e
j
u
venat
i
o
n
In
v
o
l
v
es
occa
si
ona
l
st
op
pa
ge
of
t
h
e r
u
n
n
i
n
g
so
ft
ware
, re
st
art
i
n
g i
t
a
n
d
cl
eani
n
g of t
h
e i
n
ternal state. Clea
ning
of the
syste
m
’s
internal state
m
a
y involve
fl
ushi
ng ke
rnel tables of
operat
ing system
, ga
rbage
collection and
to reinitialize
the
i
n
t
e
rnal
dat
a
s
t
ruct
u
r
es,
et
c.
Har
d
ware
re
bo
ot
i
s
wel
l
kn
o
w
n
exam
pl
e of
re
ju
ve
na
t
i
on. B
eca
use
of
t
h
e
d
i
fficu
lty in
justificatio
n
of th
e ex
t
r
a co
st
of up
-fro
n
t
d
e
velo
p
m
en
t th
e trad
ition
a
l fau
lt to
leran
ce techn
i
qu
es
are u
s
ual
l
y
av
o
i
ded i
n
t
h
e
de
v
e
l
opm
ent
of m
i
ssi
on c
r
i
t
i
cal
syste
m
s. In the
year 1996
R
o
b
e
rt J K
r
e
u
tzfel
d
, et.
al
. [1
0]
prese
n
t
e
d a m
e
t
hod
ol
o
g
y
cal
l
e
d Dat
a
Fusi
on
I
n
tegrity Pro
c
ess (DFIP) as
an
altern
ativ
e
for th
e
trad
itio
n
a
l so
ft
ware
fau
lt to
leran
ce tech
n
i
q
u
es in
ord
e
r t
o
d
eal with
h
i
gh
su
nk
co
st
o
f
dev
e
lop
m
en
t. Th
is is a
sim
p
le and effective technique
for
th
e
dev
e
lop
m
en
t o
f
fau
lt to
leran
t
m
i
ssio
n
critical syste
m
s.
A DFIP
im
pl
em
ent
a
t
i
o
n i
n
cl
u
d
es
o
n
e
reco
ver
,
o
n
e
det
ect
i
o
n
,
a
nd t
h
e re
p
o
rt
m
e
t
hod.
The
reco
ver m
e
t
h
o
d
’
s
im
pl
em
ent
a
t
i
o
n
real
i
zes bi
g
g
est
c
o
st
sa
vi
ng
i
n
DF
IP
. T
h
e m
odul
a
r
a
p
pr
oac
h
of
D
F
I
P
m
a
kes i
t
easi
e
r t
o
i
m
p
l
e
m
en
t at a
n
y ti
m
e
. Man
y
so
ft
ware
fau
l
t
to
leran
ce te
c
h
niques are a
v
ai
lable but the
relative effectiveness
o
f
d
i
fferen
t
tech
n
i
q
u
e
s
rem
a
i
n
s u
n
c
lear. S. Garn
aik,
et
al
., [11
]
in
20
13
p
r
esen
ted an ap
pro
ach
fo
r
rel
i
ab
ility
enha
ncem
ent
of so
ft
wa
re p
r
o
g
ram
s
by
m
i
nim
i
zi
ng t
h
e ove
rfl
ow er
r
o
r
s
. They
han
d
l
e
d t
h
e pr
o
b
l
e
m
of
ove
rfl
ow
beca
use of large si
ze integer i
n
put. T
h
e prese
n
ted
app
r
o
a
ch
says th
at in
stead
of storing
the larg
e
size in
teg
e
r inp
u
t
in
p
r
im
ary
in
t d
a
ta typ
e
;
sto
r
e it in
th
e lin
k
e
d
list d
a
ta stru
cture. Ev
ery n
o
d
e o
f
t
h
is lin
ked
list co
n
t
ain
s
t
w
o d
i
g
its and
p
o
i
n
t
er t
o
n
e
x
t
no
d
e
.
Au
t
h
ors p
r
esen
t tech
n
i
q
u
e
fo
r add
ition
an
d m
u
ltip
li
catio
n
o
p
e
ration
of larg
e i
n
teg
e
r inpu
t u
s
i
n
g
link
e
d list. Th
is app
r
o
ach
prov
id
es
to
leran
ce t
o
the fau
lts
o
ccurri
n
g
du
e
to
th
e
u
s
e of l
a
rg
e size
d
a
ta
to
th
e ex
ten
t
th
at as lon
g
as
th
e free m
e
m
o
ry is av
ailab
l
e for allo
cation th
ere
wo
n’t
be
any
f
a
i
l
u
re
due
t
o
us
e o
f
l
a
r
g
e si
ze
dat
a
.
Hence
,
i
t
was br
oa
dl
y
obser
v
e
d t
h
at
N-
Vers
i
on Pr
o
g
ram
m
ing
bot
h i
n
t
h
e
prese
n
ce an
d abse
nce of
d
a
ta d
i
v
e
rsity is no
t app
licab
l
e
for situatio
n
s
in
wh
ich
d
i
stin
ct m
u
ltip
le so
lu
tio
n
s
ex
ist.
Maj
o
rity v
o
ting
for
N-
Versi
on
Pr
og
r
a
m
m
i
ng m
a
y
resul
t
i
n
i
n
c
o
r
r
ect
deci
si
o
n
whe
n
m
a
jori
t
y
of
out
put
s a
r
e i
n
co
rrect
.
D
e
si
gn
di
ve
rsi
t
y
m
a
y
not
be ve
ry
eff
ect
i
v
e i
f
sim
i
l
a
r ki
n
d
of fa
ul
t
s
get
det
ect
ed i
n
va
ri
o
u
s ve
rsi
ons
. Im
pl
em
ent
a
t
i
o
n
of e
ffect
i
v
e
de
si
gn
di
ve
rsi
t
y
m
a
y
i
n
cur hi
g
h
cost
due t
o
t
h
e de
vel
o
pm
ent
of m
a
ny
vers
i
ons
. He
nce i
t
i
s
not
a
cost
effect
i
v
e
app
r
oach f
o
r real
t
i
m
e
sy
stem
s. Tho
u
g
h
im
pl
em
ent
a
t
i
o
n of
dat
a
di
v
e
rsi
t
y
based
N-C
opy
Pro
g
r
am
m
i
ng usi
n
g dat
a
re
-e
xp
ressi
on al
g
o
r
i
t
h
m
i
s
a cost
e
ffectiv
e ap
proach
, yet it is n
o
t
v
e
ry
u
s
efu
l
as it is
d
i
fficu
lt to
ob
t
a
in
an
effective d
a
ta re-expressio
n
al
go
r
ith
m
for
vari
o
u
s r
eal tim
e
sy
s
t
e
m
. The effectivene
ss
of
vari
ous
faul
t
t
o
l
e
ra
nce t
ech
n
i
ques
vari
es w
i
t
h
vari
at
i
o
ns i
n
sy
st
em
s. Th
eref
ore t
h
e
r
e i
s
no
uni
versal
faul
t
to
leran
ce techn
i
qu
e wh
ich
i
s
effectiv
e
for all real
tim
e
syste
m
s. There is no m
e
thodol
ogy to a
sse
ss the
rel
a
t
i
v
e effect
i
v
ene
ss o
f
di
ffe
rent
fa
ul
t
t
o
l
e
r
a
nce t
ech
ni
q
u
e
s
. R
eco
very
B
l
ock Tec
h
ni
q
u
e
i
s
not
s
u
i
t
a
b
l
e for
real ti
m
e
syste
m
s
d
u
e
to
it’s serial n
a
tu
re o
f
ex
ecu
tion
.
Fro
m
th
e su
rvey, it h
a
s
been found th
at there are
several
wo
rk
s
conce
r
ned
wi
t
h
achi
e
vi
n
g
fa
ul
t
t
o
l
e
rance i
n
t
h
e pre
s
ence
o
f
va
ri
o
u
s f
u
nct
i
onal
an
d
ope
r
a
t
i
onal
err
o
rs
b
u
t
m
a
ny
of t
h
em
di
d not
a
d
dress
t
h
e
pr
o
b
l
e
m
concerne
d
wi
t
h
t
i
m
i
ng
b
ugs
w
h
i
c
h i
s
a
n
i
m
port
a
nt
i
ssu
e
in real tim
e syste
m
s. As
for
real tim
e
syste
m
s,
m
a
ny
ti
me
s it b
e
co
m
e
s a
n
ecessity fo
r
a g
i
v
e
n
serv
ice to
b
e
d
e
liv
ered
within
th
e
sp
eci
fied
tim
e d
ead
line. Th
erefore t
h
e m
a
in
ob
j
e
ctiv
e is to
d
e
v
e
lo
p
a v
e
rsatile co
st
effective a
p
proach in t
h
e
presence
of
timin
g
bug
s
f
o
r
pro
v
i
d
i
ng
f
a
u
lt
tolera
nce t
o
real tim
e soft
ware
applications.
Hence
,
a cost effective approac
h
in
t
h
e prese
n
ce o
f
t
i
m
i
ng bu
gs ha
s been de
vel
o
ped f
o
r
p
r
ov
id
ing
fau
l
t
to
leran
ce to real
ti
m
e
so
ftware app
licatio
n
s
.
3.
THE PROPOSED
APPROACH
In acc
or
dan
c
e
wi
t
h
t
h
e pr
o
pos
ed a
p
p
r
oac
h
[
12]
, t
h
e e
n
t
i
r
e so
ft
war
e
sy
st
em
i
s
divi
de
d i
n
t
o
subsystem
s
. Risk assessm
ent is done
on
various tas
k
s
for
each s
ubsyste
m to iden
tify thos
e tasks
whi
c
h are
m
o
re critical.
From
the task
depe
ndency a
n
alysis dead
line
is obtaine
d
for each s
u
bsystem
.
W
h
e
n
a s
ubsyste
m
i
s
execut
e
d
,
b
o
t
h t
h
e ori
g
i
n
al
pr
o
g
ram
and t
r
ai
ned ne
u
r
al
n
e
t
w
o
r
k (
T
N
N
)
m
odul
e i
s
execut
e
d si
m
u
l
t
a
neousl
y
for th
e critical task
s of th
at sub
s
ystem
.
If th
ere is an
y
in
com
p
le
te/
m
issin
g
d
a
ta th
en
th
e
p
r
ed
ictio
n
m
o
du
le is
u
s
ed
to
prov
ide th
e m
i
ssin
g
data. Th
ere is a
co
nstan
t
p
o
lling
to ch
eck
wh
eth
e
r t
h
e
o
u
t
p
u
t
is av
ailab
l
e from
th
e
o
r
i
g
in
al
p
r
og
ram
.
If th
e
ou
tput is av
ailab
l
e
fro
m
th
e o
r
i
g
in
al b
l
o
c
k
b
e
fore
th
e tim
e (
deadl
i
n
e –
k
)
(w
her
e
k
is
t
h
e reaso
n
a
b
l
e
am
ount
of t
i
m
e requi
re
d f
o
r com
p
l
e
t
i
n
g
t
h
e operat
i
on
s conce
r
ne
d
wi
t
h
del
i
v
e
r
i
n
g t
h
e
app
r
op
ri
at
e
o
u
t
put
fo
r su
bse
que
nt
pr
ocessi
ng
a
n
d
d
e
ad
line
co
rrespon
ds to th
e
d
e
ad
lin
e tim
e fo
r the task
co
m
p
letio
n
)
then
it is fo
rward
e
d
furth
e
r fo
r su
b
s
equ
e
n
t
p
r
o
cessing
else if th
e ou
tpu
t
is av
ailab
l
e from th
e
TNN
bl
oc
k t
h
e
n
i
t
i
s
fo
rwa
r
de
d
fo
r s
u
bseq
ue
nt
p
r
ocessi
n
g
.
Ot
he
rwi
s
e,
t
h
e
sub
s
y
s
t
e
m
m
o
ves t
o
t
h
e
safe
m
ode
and
sen
d
s a
p
p
r
op
ri
at
e si
gn
al
f
o
r s
u
bse
que
nt
pr
ocessi
ng
. Th
e del
a
y
i
n
p
r
o
d
u
ci
n
g
t
h
e
out
p
u
t
fr
om
t
h
e ori
g
i
n
al
p
r
og
ram
m
a
y
d
e
p
e
nd
o
n
lo
t
o
f
fact
o
r
s lik
e
task
d
e
p
e
nd
ency, reso
urce
sharing
,
un
av
ailab
ility o
f
t
h
e req
u
i
red
data etc. T
h
e
detailed steps c
o
ncerne
d
with
t
h
is ap
pro
a
ch
is sp
ecified
in al
g
o
rith
m
fa
ult_
t
o
l (system)
The proposed algorithm
is
speci
fied
in th
e
fo
llo
wi
n
g
step
s:
-
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E
V
o
l
.
6,
No
. 3,
J
u
ne 2
0
1
6
:
10
3
1
– 10
37
1
034
Algo
rith
m
faul
t_
tol
(sy
s
tem
)
{
Step 1
:
Div
i
d
e
th
e real tim
e sy
ste
m
in
to
sub
s
yste
m
s
b
a
sed
on
task
s.
Step 2
:
Filter th
e task
b
a
sed on
risk
assessmen
t.
Step 3
:
Ob
tain
d
ead
lin
e fo
r each
sub
system
b
a
se
d
on
t
h
e t
a
sk
de
pen
d
e
n
cy
anal
y
s
i
s
.
Step 4
:
For eac
h s
u
b system
repeat step 5 to
step
10.
Step 5
:
Start the ti
m
e
r.
Step 6
:
If th
e i
n
pu
t
d
a
ta set is co
m
p
lete th
en
p
a
ss t
h
e i
n
p
u
t
t
o
T
r
ai
ne
d
Neu
r
al
Net
w
or
k
(T
NN
)
bl
oc
k a
n
d
ori
g
i
n
al
pr
og
ra
m
.
El
se pass i
n
c
o
m
p
l
e
t
e
i
nput
t
o
pre
d
i
c
t
i
on m
odul
e a
nd t
h
en
pass p
r
e
d
i
c
t
e
d
com
p
l
e
t
e
i
nput
dat
a
t
o
TN
N
bl
oc
k a
n
d
o
r
i
g
i
n
al
p
r
og
ram
.
Step 7
:
St
art
e
x
ecut
i
o
n
o
f
T
N
N
bl
oc
k a
n
d
o
r
i
g
i
n
al
p
r
og
ram
si
m
u
l
t
a
neousl
y
.
Step 8
: If
ti
m
e
r
< (Dead
lin
e –
k)
Step 8.
1
:
If th
e ou
tpu
t
of t
h
e
orig
in
al
p
r
og
ram
is av
a
i
l
a
bl
e t
h
en
f
o
r
w
a
r
d
i
t
fo
r s
u
b
s
eq
ue
nt
pr
ocessi
ng
Else go to step
8.1.
Step 9:
If tim
e
r
> = (Dead
line –
k)
Step 9.
1
:
If
t
h
e
o
u
t
p
ut
f
r
om
t
h
e Trai
ne
d
Ne
ur
al
Net
w
or
k
bl
o
c
k i
s
a
v
ai
l
a
bl
e
t
h
en
f
o
r
w
ar
d i
t
f
o
r s
u
bse
que
nt
pr
ocessi
ng
Else go to step
10.
Step 10
:
M
o
ve
t
h
e s
u
bsy
s
t
e
m
t
o
t
h
e
safe m
o
d
e
an
d se
n
d
a
p
p
r
o
p
ri
at
e si
gnal
fo
r s
u
b
s
eq
ue
nt
pr
ocessi
ng
.
}
Whe
r
e,
k
i
s
t
h
e reaso
n
abl
e
a
m
ount
of t
i
m
e
requi
red f
o
r com
p
l
e
t
i
ng t
h
e ope
rat
i
o
ns co
n
cerne
d wi
t
h
del
i
v
eri
ng t
h
e
app
r
op
ri
at
e out
put
f
o
r s
u
b
s
eq
uent
p
r
oce
ssing
d
e
ad
lin
e correspo
nd
s to
th
e
d
ead
lin
e tim
e
fo
r th
e
task
co
m
p
letio
n
.
4.
IMPLEME
N
TATION DE
TAILS
Prese
n
t
l
y
, wo
r
k
i
s
i
n
pr
o
g
re
ss fo
r t
h
e i
m
pl
em
ent
a
t
i
on of
t
h
e Trai
ned
Neu
r
al
Net
w
o
r
k
bl
oc
k
f
o
r
v
a
ri
o
u
s
fu
n
c
tion
s
.
As alread
y
stated
th
at fo
r
ach
iev
i
ng
t
h
e fau
lt to
leran
c
e in
real tim
e sys
t
e
m
s, it is n
ecessary
to
prov
id
e temp
oral correct
n
e
ss alo
n
g
with
p
r
ov
id
ing
f
u
nc
t
i
onal
co
rrect
n
e
ss. T
h
i
s
w
o
r
k
has f
o
c
u
se
d o
n
t
h
e
aspect of providing t
h
e tem
p
oral c
o
rrectnes
s
to the r
eal time softwa
re s
y
ste
m
s. Som
e
basic operations like
ad
d
ition
an
d su
b
t
ractio
n u
s
i
n
g
th
e n
e
ural network h
a
d
alread
y b
e
en
im
p
l
em
en
ted
for 32
b
it in
teg
e
rs. Fo
r
t
r
ai
ni
n
g
, a
f
u
l
l
ad
der
l
o
gi
c
h
a
s bee
n
use
d
.
The i
n
p
u
t
dat
a
use
d
f
o
r
t
r
ai
n
i
ng
t
h
e
ne
ural
net
w
or
k i
s
s
h
o
w
n
i
n
Tabl
e 1.
Tabl
e
1. B
i
nar
y
i
n
p
u
t
com
b
i
n
at
i
on
fo
r
ful
l
a
dde
r
Cin A
B
Su
m
Cout
0 0
0
0
0
0 0
1
1
0
0 1
0
1
0
0 1
1
0
1
1 0
0
1
0
1 0
1
0
1
1 1
0
0
1
1 1
1
1
1
Fo
r t
r
ain
i
ng
the n
e
ural n
e
twork
for th
is b
a
sic ad
d
ition
op
eratio
n
b
a
ck
p
r
op
ag
ation
was used
.
At the
input layer the
r
e are
2 nodes
, at hidd
en lay
e
r there a
r
e
4 nodes a
nd at t
h
e output layer finally there
are 2
no
des
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Review of
Soft
w
are F
a
ult-Tol
e
rance Methods for
Reliab
ility Enhance
m
e
n
t of Re
al-Ti
m
e .... (A
njus
hi Ver
m
a)
1
035
Fi
gu
re 1.
Ne
ur
al
net
w
or
k
use
d
fo
r t
h
e
i
m
pl
em
ent
a
t
i
o
n
Th
e
fin
a
l
weigh
t
m
a
trix
o
f
the edg
e
s
b
e
tween
v
a
ri
o
u
s
layers is as
fo
llows:
Tabl
e
2.
Fi
nal
l
earne
d
wei
g
ht
m
a
t
r
i
x
of
i
n
p
u
t
t
o
hi
d
d
en
l
a
y
e
r e
dges
1 2 3
4
1
-
2
.
25661
097
601
5.
5340
364
438
9
-
4
.
31645
140
220
6.
0342
393
365
9
2 -
2
.
58656
210
241
6.
7137
172
161
1
4.
5934
948
072
4
2.
1564
776
527
7
3 -
3
.
42512
067
886
5.
8292
800
477
4
5.
9865
453
818
9
-
2
.
02944
103
020
Tabl
e
3.
Fi
nal
l
earne
d
wei
g
ht
m
a
t
r
i
x
of
hi
d
d
e
n t
o
o
u
t
p
ut
l
a
y
e
r ed
ges
1
2
1 -
1
.
15939
605
063
6.
3287
400
729
5
2 10.
481
796
167
72
6.
3382
927
274
5
3 9.
4549
433
746
1
1.
1567
151
247
2
4 9.
4995
642
449
7
2.
7487
704
058
3
Aft
e
r
succe
ssf
ul
t
r
ai
ni
n
g
o
f
t
h
i
s
ne
u
r
al
net
w
o
r
k
bl
ock
i
t
has
been
t
e
st
ed f
o
r va
ri
o
u
s i
n
t
e
ge
r i
n
put
s
with
in
t
h
e
range. So
m
e
o
f
th
e
resu
lts are as fo
llo
ws:
dem
o
(1
50
,-
60
0
,
F_
W_
I
H
,F
_
W
_H
O,
F_B
_
H
,
F
_
B
_
O)
ans =
-450
dem
o
(2
00
,4
0
0
,
F
_
W
_
I
H
,
F
_
W
_
H
O
,
F
_B
_H
,F
_B
_
O
)
an
s = 600
It
i
s
fo
u
nd t
o
per
f
o
r
m
t
h
e operat
i
o
n c
o
r
r
e
c
t
l
y
. The p
r
ed
i
c
t
i
on bl
ock i
s
bei
n
g i
m
pl
em
ent
e
d
usi
n
g
M
a
rk
ov
m
odel
an
d
vari
o
u
s
ot
her
st
eps
o
f
t
h
i
s
ap
pr
oac
h
a
r
e
bei
n
g a
p
pl
i
e
d t
o
vari
ou
s r
eal
-t
im
e appl
i
cat
i
ons t
o
o
b
t
ain th
e
results.
5.
DIS
C
USSI
ONS
Fai
l
u
res
due t
o
t
i
m
i
ng b
u
g
s
m
a
y
depen
d
on se
ve
ral
fact
ors
base
d o
n
t
a
sk d
e
pe
n
d
enc
i
es, res
o
u
r
ce
sh
ar
i
n
g,
o
p
e
r
a
t
i
o
n
a
l
p
r
o
f
ile, etc. I
f
it is co
nsid
er
ed
th
at th
e f
a
ilu
r
e
s caused
b
y
ti
m
i
n
g
bug
s
o
ccur
r
a
ndomly
,
th
en
th
e
failu
re rate d
u
e
t
o
ti
min
g
b
u
g
s
is constan
t
. Let th
e co
n
s
tan
t
failure rate d
u
e
t
o
ti
m
i
n
g
bu
g
s
b
e
k
. I
f
the
failu
re rate
o
f
th
e act
u
a
l so
ftware
(in th
e
p
r
esen
ce
of timin
g
bug
s) is
z
(t)
. Th
en the failure rate
o
f
fau
lt-
to
ler
a
n
t
pr
og
ra
m
s
(
a
vo
id
i
n
g ti
m
i
n
g
bug
s b
a
sed
on
t
h
e pr
opo
sed
ap
pro
ach)
is
z(t) – k
. If
z(
t)
is an
ex
pon
en
tially d
ecr
easi
n
g functio
n
[
1
1
]
(du
r
in
g
t
h
e testing/d
e
bu
gg
ing phase)
t
h
en th
e
p
l
o
t
s i
n
Fi
g
u
r
e
2 for
act
ual
and fa
ul
t
-
t
o
l
e
ra
nt
pr
og
ram
s
are obt
ai
ned
.
If
z(
t)
is a lin
early d
ecreasing
fun
c
tion
(durin
g
t
h
e
testin
g
/
d
e
b
uggin
g
ph
ase) th
en th
e
p
l
o
t
s i
n
Fi
g
u
r
e
3
f
o
r
act
ual an
d f
a
u
lt-
to
l
e
r
a
n
t
p
r
og
r
a
m
s
ar
e
ob
tain
ed.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E
V
o
l
.
6,
No
. 3,
J
u
ne 2
0
1
6
:
10
3
1
– 10
37
1
036
Hen
c
e, it is ev
id
en
t th
at th
e failu
re rate o
f
th
e fau
lt-t
o
leran
t
p
r
o
g
ram
s
(o
b
t
ain
e
d
fro
m
t
h
e p
r
op
osed
approach) are
always less than
th
at of th
e orig
in
al p
r
og
ram
s
wh
en
timin
g
b
u
g
s
are p
r
esen
t. Th
e
failu
re rates
are equal whe
n
the
r
e are
no
timing bugs.
Sin
ce, it is o
b
s
erv
e
d
th
at th
e
failu
re rate is less for fau
lt-tolerant
pr
o
g
ram
s
(obt
ai
ned f
r
o
m
t
h
e pr
op
ose
d
ap
p
r
oac
h
) as c
o
m
p
are
d
t
o
t
h
e
o
r
i
g
i
n
al
p
r
og
ra
m
i
n
t
h
e prese
n
ce o
f
ti
min
g
bu
g
s
.
Hen
ce, it can b
e
in
ferred th
at t
h
e
reliab
ility
v
a
lu
e o
f
th
e g
i
ven
o
r
i
g
in
al p
r
og
ram
is
enh
a
nced
i
n
t
h
e pres
ence
o
f
t
i
m
i
ng bu
gs
by
usi
n
g t
h
e p
r
o
p
o
sed a
p
p
r
o
ach t
o
o
b
t
a
i
n
t
h
e fa
ul
t
-
t
o
l
e
ra
nt
pr
o
g
ram
.
Tabl
e 4,
sh
own
th
e com
p
ariso
n
of the p
r
o
p
o
s
ed
app
r
o
a
ch
with
o
t
h
e
r ap
pro
a
ch
es fo
r th
ei
r su
itab
ility with
resp
ect to
real-tim
e syste
m
s.
Tab
l
e
4
.
C
o
m
p
arison
o
f
th
e suitab
ility o
f
v
a
rio
u
s
ap
pro
a
ch
es with
resp
ect to real-tim
e syste
m
s
Techniques
Suitability for
real ti
m
e
syste
m
s
Reasons
N-
Ver
s
ion Pr
ogr
am
m
i
ng
Poor
o
Not suitable when distinct
m
u
ltip
le solutions
exist. Majority
voting
for
N-
Version Pr
ogr
am
m
i
n
g
m
a
y
r
e
sult in incor
r
ect
decision when
m
a
j
o
rity of
outputs are incorrect.
o
Design diver
s
ity
m
a
y
not be ver
y
effective i
f
sim
ilar
kind
of
faults get detected in var
i
ous ver
s
ions
o
T
h
e decision
m
e
chanis
m
m
a
y
not be adequate enoug
h t
o
handle
tim
i
ng bugs for
pr
ovidin
g
tem
poral cor
r
ectness.
Recover
y
Blocks[9]
Poor
o
I
t
m
a
y not be suitable for
r
eal ti
m
e
s
y
stem
s due to it’
s
ser
i
al
natur
e
of executio
n.
o
No
m
echanis
m
s
to handle tim
i
ng bugs for
pr
ovidin
g
tem
poral
correctness.
N-Self
Checking
Pr
ogr
am
m
i
ng
Fair
o
Design diver
s
ity
m
a
y
not be very
effective if si
m
ila
r kind of
faults get detected in var
i
ous ver
s
ions
o
No
m
echanis
m
s
to handle tim
i
ng bugs for
pr
ovidin
g
tem
poral
correctness.
N-
Copy
Pr
ogr
a
m
m
i
ng
Fair
o
Since,
it is
difficult
to obtain an effective data r
e
-
e
xpr
ession
algor
ith
m
for
var
i
ous r
eal ti
m
e
s
y
stem
s
.
o
No
m
echanis
m
s
to handle tim
i
ng bugs for
pr
ovidin
g
tem
poral
correctness.
Retr
y Blocks
Poor
o
I
t
m
a
y not be suitable for
r
eal ti
m
e
s
y
stem
s due to it’
s
ser
i
al
natur
e
of executio
n
o
No
m
echanis
m
s
to handle tim
i
ng bugs for
pr
ovidin
g
tem
poral
correctness.
Pr
oposed appr
oach
[14]
Good
o
Provides
m
echanis
m
to handle ti
m
i
n
g
bugs a
nd
unavail
ability of
data,
ther
eby
pr
o
v
iding tem
por
al cor
r
ectness,
in addition to
pr
ovidin
g
functi
on
al cor
r
e
ctness for
real ti
m
e
softwar
e
s
y
stem
s
.
From
, Tabl
e
4,
i
t
i
s
fo
u
nd t
h
at
t
h
e p
r
o
p
o
se
d a
p
p
r
oac
h
i
s
f
o
u
n
d
t
o
be m
o
re s
u
i
t
a
bl
e as c
o
m
p
ared t
o
othe
r a
p
proac
h
es for real tim
e
softwa
re syste
m
s.
Fig
u
re
2
.
Failure rates fo
r expo
n
e
n
tiallyd
ecreasin
g z(t)
Fig
u
re
3
.
Failure rates fo
r lin
early d
ecreasing z(t)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Review of
Soft
w
are F
a
ult-Tol
e
rance Methods for
Reliab
ility Enhance
m
e
n
t of Re
al-Ti
m
e .... (A
njus
hi Ver
m
a)
1
037
6.
CO
NCL
USI
O
N
A real
-tim
e syste
m
should
process i
n
form
ation a
n
d
produc
e a res
p
onse
within a s
p
ecifie
d
tim
e. Real
ti
m
e
sys
t
e
m
s
are ti
m
e
critic
al an
d
th
ei
r co
rrectn
e
ss
d
e
pen
d
s on
b
o
t
h
th
e correct
n
e
ss o
f
ou
tpu
t
and
th
eir
tim
e
liness. In
real tim
e
syste
m
s failure m
a
y cause
ve
ry
se
vere
co
nse
q
ue
nces.
Si
nce
i
t
i
s
n
o
t
al
way
s
p
o
ssi
bl
e
to
en
su
re th
e
d
e
v
e
l
o
p
m
en
t of h
i
gh
ly reliable real
time s
y
ste
m
s hence they are prime candi
dates for the
i
n
cl
usi
o
n
of
fa
ul
t
t
o
l
e
ra
nce t
e
chni
que
s.
In t
h
i
s
pape
r,
re
vi
ews
of t
h
e e
x
i
s
t
i
ng a
p
pr
oac
h
es
fr
om
t
h
e pers
pect
i
v
e
of
real tim
e
s
y
ste
m
s to analyze th
e sho
r
t
c
om
i
ngs of t
h
e
s
e app
r
oaches
and
fi
nal
l
y
p
r
ocee
ds t
o
pre
s
ent
a
v
e
rsatile an
d
co
st effectiv
e ap
pro
ach
in
th
e p
r
esen
ce
o
f
timin
g
b
u
g
s
fo
r p
r
ov
id
i
n
g
fault to
leran
ce, thereby
en
h
a
n
c
ing
the reliab
ility o
f
th
e real ti
me so
ft
ware app
licatio
n
s
.
The p
r
op
o
s
ed
ap
pro
ach
n
e
ed
s to
b
e
im
pl
em
ent
e
d f
o
r
va
ri
o
u
s
real
t
i
m
e
soft
wa
re
ap
pl
i
cat
i
ons.
Thi
s
a
p
p
r
oach
i
s
hel
p
ful
i
n
pr
ovi
di
n
g
t
h
e
t
e
m
poral
cor
r
ect
ness
f
o
r
real
t
i
m
e
sy
st
em
s i
n
t
h
e presence of t
i
m
i
ng b
ugs t
h
e
r
e
b
y
pr
ovi
di
n
g
faul
t
t
o
l
e
ra
nce an
d
m
a
ki
n
g
th
e app
licatio
n m
o
re reliab
l
e.
ACKNOWLE
DGE
M
ENTS
Th
e au
tho
r
s
wo
u
l
d
lik
e t
o
t
h
an
k all th
e staffs of SC
& SS, JNU fo
r th
eir sup
port b
e
h
i
nd
t
h
is work
.
REFERE
NC
ES
[1]
H.
Kope
tz
,
“Re
a
l-Time
S
y
ste
m
s: De
sign Princ
i
ples for Distributed Em
bedded Applications,”
Will
ey
, 2nd edition
,
2006.
[2]
Anderson T. and Knight J. C., “A Framework
for
Software Fault Tole
ranc
e
in Real-T
im
e S
y
stem
s,”
IE
EE
Transactions
on Software Engineering
, vo
l/issue:
9(3), pp
. 355-36
4, 1983
.
[3]
X. Zaip
eng,
et al
.
,
“A Study
O
f
Softwa
re
Fa
ult Toler
a
nce Techniques,”
1415
Engineering Drive, Mad
i
son WI
53706 USA
, 199
8s.
[4]
L. Chen
and A.
Avizien
i
s
,
“
N
-Vers
i
on Programming: A Fault-Toler
a
nce Appr
oach to
Rel
i
abi
lit
y of S
o
ftwa
r
e
Operation
,
”
Pr
o
ceed
ings
of
FT
CS- 8,
T
oulous
e,
Fr
ance
, 1978, p
p
. 3–9
.
[5]
J.
J.
Horning,
et al.
, “A program structure for er
ro
r detection and r
ecover
y
,” in
O
p
e
r
. Sys
t
.,
Pr
oc. Int
.
Sym
p
(
L
ectur
e
Notes in
Computer Science)
, vol.
16, E. Gele
nb
e a
nd C. Ka
iser,
Ed
s. Berlin: Spring
e
r-Verlag
, 1974, pp. 171-187.
[6]
J. Gray
, “Why
Do Computers
Stop a
nd What Can Be Done About It?
”
Proc. Fifth Symp. Reliabi
l
ity in Distribut
e
d
Software and
Database Systems,
Jan,
1986
, pp
. 3
-
12.
[7]
A. Avizienis a
nd J. P.
J. Kell
y, “
F
ault Toler
a
nce
by
Design Diversity
:
C
oncepts and Experiments,”
IE
EE
Computer
, vol/is
sue: 17(8), pp
. 6
7–80, 1984
.
[8]
P. E. Ammann and J. C.
Knight, “Data Diversity: An Appro
ach to Software Fault Toleran
c
e,”
I
E
EE Transactions
on Computers
, v
o
l/issue: 3
7
(4), p
p
. 418–416
, 198
8.
[9]
K. Kanoun,
et al
., “
R
el
iabi
lit
y
growth of fault
toler
a
nt softwar
e
,”
I
EEE T
r
ansactions on Re
li
abilit
y
, vol/issue:
42(2), 1993
.
[10]
J. K. Rob
e
rt
and
R.
E. Neese, “
A
Methodolog
y
For Cost Effecti
v
e Softwa
re Fau
lt to
ler
a
nce For
Mission-Critical
S
y
ste
m
s,
”
15
th
AIAA
/IEE
E Dig
i
tal
Avion
i
cs S
y
st
ems conferen
ce
,
Atlanta
,
GA
, 199
6, pp
. 19-24
.
[11]
S.
Ga
rna
i
k,
et
al
.,
“
R
eli
a
bili
t
y
Enhan
cem
ent
of S
o
ftware
b
y
Minimizing
th
e Overflow
Err
o
rs,”
Internatio
nal
Journal of Systems Assurance E
ngineering
and
Management, Sp
ringer
, vol/issue: 5(4), pp. 724-7
30, 2014
.
[12]
Ghartaan A.
and
Gay
e
n
T., “Anal
y
sis and propos
ition of fau
lt-to
l
e
rance m
odel fo
r real t
i
m
e
software s
y
st
em
s,” in
the pr
oceed
ings
of Inter
nationa
l Confer
enc
e
on
Compu
ting, Communication, E
l
ectr
i
cal,
Ele
c
tr
o
n
ics
,
Dev
i
ces
&
Signal
Processin
g
, India
,
Lankap
alli, India
, pp
. 3
53-357,
Discovery
,
vol/issue: 44(
202), pp
62-67,
2015.
Evaluation Warning : The document was created with Spire.PDF for Python.