You are on page 1of 5

Faul

tTol
erance
Faul
tToleranceinsof
twareisaphenomenonwherethesoft
wareiscapabl
eoffi
xingi
tsel
for
conti
nui
ngt henormaloperati
onsi
nt heoccurr
enceofanyglit
chesorerrori
nthesy st
em,
provi
dedthatf ul
lcover
ageont hefunct
ional
i
tyismaint
ainedasspecifi
edintherequir
ed
documentati
on.
Soft
warefaul
ttol
eranceistheabili
tyforsoft
waretodetectandrecov
erfr
om afaul
tthatis
happeni
ngorhasalreadyhappenedineithert
hesoft
wareorhardwareint
hesyst
em i
nwhi ch
thesoft
warei
srunni
ngt opr
ovideservi
cebythespeci
fi
cat
ion.

Ther easonsbehindthesefaul
tsinthesoftwaresystem canbeaf aul
tfrom wi
thin,f
rom other
i
ntegratedsy st
ems,from t
hedownst ream appl
icat
ionorf r
om anyotherexter
nalaspectslike
thesy stem hardwar
e,network,etc.Thisisoneoft hefactor
sbasedonwhi chsof twareis
esti
mat edtobeaqual i
tyoneornot.Henceitisi
mpor tantt
hatever
ysoftwareprogram consi
sts
offaulttol
erance.
.
Faul
tTol
eranceTechni
ques
AFaul tinanysof t
waresy stem,usually
,happensduet ot hegapsl ef
tunnot i
ceddur i
ngt he
designphase.Basedont his,thefaulttoler
ancetechni
quesar eident
if
iedintotwodi ff
erent
groups,thatis,theSingl
eVer sionTechniqueandt heMul t
i-Versi
onTechnique.Therecanbe
pl
ent yoftechniquesimplement edundereachoft hesecategories,andafewoft hetechniques
oftenusedbyt heprogrammer sare,
Si
ngl
eVer
sionTechni
que
 Sof
twar
eSt
ruct
ureandAct
ions
 Er
rorDet
ect
ion
 Except
ionHandl
i
ng
 Checkpoi
ntandRest
art
 Pr
ocessPai
rs
 Dat
aDi
ver
sit
y
Mul
ti
-Ver
sionTechni
que
 Recov
eryBl
ocks
 N–Ver
sionPr
ogr
ammi
ng
 NSel
f–Checki
ngPr
ogr
ammi
ng
 ConsensusRecov
eryBl
ocks

Si
ngl
eVer
sionTechni
que
1.Sof
twar
eSt
ruct
ureandAct
ions
Whent hesof twar esy st
em i
sonesi ngl
eblockofcode,itisl
ogicall
ymor ev
ulnerabl
etofail
ure.
Because,whenonet inyer
roroccursintheprogram,thewholesy stem wi
llbebroughtdown.
Hence,itiscr ucialforthesoft
waresy st
em shouldbestruct
uredinamodul arform,wherethe
funct
ionali
tyiscov eredinseparatemodules.I
nt hecaseoff ail
ure,eachmoduleshouldhold
speci
ficinstructionsonhow t ohandl eitandlettheothermodul esrunperusual,i
nsteadof
passi
ngont hef ail
uref r
om moduletomodule.
2.Er
rorDet
ect
ion
ErrorDetect
ioni
saf aul
ttoler
ancet echniquewheretheprogram l
ocateseveryincidenceoferror
i
nt hesy st
em.Thistechniqueispracticall
yimplementedusingtwoat t
ri
butes,namel y,thesel
f-
protect
ionandt heself-
checking.TheSel f
-Prot
ecti
onattr
ibuteofer r
ordetectioni susedf or
spotti
ng the er
ror
si nt he exter
nalmodul es,whereasthe Self
-Checki
ng at t
ribute oferror
detecti
onisusedforspotti
ngtheer ror
si ntheint
ernalmodule.
3.Except
ionHandl
ing
Excepti
onHandli
ngisat echni
queusedforredi
recti
ngtheexecut
ionf
lowtowar
dst herouteto
recover
ywheneveraner r
oroccursinthenormalfuncti
onalf
low.Asapartoffaultt
olerance,
thi
sact i
vi
tyisperf
ormedundert hr
eediffer
entsoftwarecomponent
s,suchast heInterf
ace
Excepti
on,t
heLocalExcepti
onandtheFail
ureExcepti
on.
4.Checkpoi
ntandRest
art
Thisisoneoft hecommonl yused recuperationmet hodsf orthesi ngl
ev ersionsof tware
systems.TheCheckpoi ntandRestartfaul
ttolerancetechniquecanbeusedf ortheev ent
slike
run-
timeexceptions,thati s,amal f
uncti
ont akesplacedur ing t
her un-t
imeand whent he
executi
oniscompl et
et hereisnorecordoft heer r
orhappening.Forthi
scase,t hepr ogrammer
canplacecheckpointsi
nt heprogr
am andi nstr
uctt hepr
ogram t or
estar
timmedi atelyri
ghtfrom
theoccurr
enceoft heerror.
5.Pr
ocessPai
rs
ProcessPairt
echniqueisamet hodofusingsamesoftwarei
ntwodif
fer
enthardwareunits,and
vali
dati
ng the f
unctionaldi
ff
erences i
n orderto capt
urethe f
aul
tyareas.Thist echnique
functi
onsontopoft hecheckpointandrest
artt
echni
que,asthesi
mil
archeckpoi
ntsandr estar
t
i
nstructi
onsareplacedinbot
ht hesyst
ems.
6.Dat
aDi
ver
sit
y
DataDiver
sityt
echniqueistypi
call
yaprocesswheretheprogrammerpassesasetofi nputdat
a,
andplacescheckpointsfordetect
ingthesli
ppage.Thecommonl yusedDat aDiver
sitymodels
are‘
InputDataRe-Expressi
on’model,‘
I
nputDataRe-Expr
essionwithPost
-Execut
ionAdjustment’
model,and‘Re-
Expressionvi
aDecomposi t
ionandRecombination’
model.

Mul
ti
-Ver
sionTechni
que
1.Recov
eryBl
ocks
Recov eryBlocktechni
queformult
ipl
ev er
sionsoftwareFaul
tToler
anceinvol
vest hecheckpoint
andr estartmet hod,wher
et hecheckpoi
nt sareplacedbefor
et hefaultoccurrence,andthe
system isinstructedtomoveont onextv er
siontoconti
nuethef l
ow.Itiscarr
iedouti nthr
ee
areas,thatis,t
hemai nmodule,
theacceptancetest
s,andtheswapmodul e.
Recover
y-Bl
ock-Scheme–
Therecover
yblockschemeconsist
softhreeel
ements:pr
imarymodule,acceptancet
est
s,and
al
ter
natemodulesforagiv
entask.Thesi
mplestschemeoftherecov
eryblockisasfol
l
ows:

Ensur
eT
ByP
El
sebyQ1
El
sebyQ2
……….
El
sebyQn-
1
El
seEr
ror
WhereTi sanacceptancet
estcondi ti
ont hati
sexpectedt obemetbysuccessf ulexecuti
onof
ei
therthepr
imarymodulePort healternatemodulesQ1, Q2,...,
Qn- 1.
Theprocessbeginswhent heout putoft heprimarymodul eist estedforaccept abi
li
ty.Ifthe
accept
ancetestdet
erminesthattheout putofthepri
mar ymodul eisnotaccept able,i
trecovers
orroll
sbackt hestateofthesy stem bef oretheprimarymodul ei sexecuted.Itallowst he
secondmodul eQ1,t oexecute.Theaccept ancetesti srepeatedt o checkthesuccessf ul
execut
ionofmoduleQ1.Ifi
tfail
s,thenmodul eQ2i sexecuted,et
c.
Theal t
ernatemodul esar eidentifi
edbyt hekeywor ds“ elseby”Whenal lalter
natemodulesar e
exhausted,therecov erybl ocki tsel
fi sconsideredt ohav efai
ledandt hefinalkeywords“else
error
”declaresthefact .Inot herwor ds,whenal lmodul esexecut eandnonepr oduceaccept
abl e
outputs,thenthesy stem f alls.A r el
iabil
i
tyopt i
mi zationmodelhasbeenst udi
edbyPham
(1989b)t o det
ermi ne t he opt i
malnumberofmodul es inar ecoverybl ock scheme that
minimizesthetotalsy stem costgi vent hereli
abil
it
yoft hei ndi
vi
dualmodul es.

2.N–Ver
sionPr
ogr
ammi
ng
TheN–Ver si
onprogrammi ngtechni
quef ort
hemulti–ver
sionfaul
ttol
eranceisthecommonl y
usedmet hodwhent hethereisapr ovi
sionfortest
ingmult
ipl
ecodeedi ti
ons.Therecoveryis
madef r
om executi
ngal ltheversi
ons,andcompar ingt
heoutputsfrom eachoft hever
sions.
Thist
echniqueal
soinv ol
vestheacceptancetestf
low.

N-v
ersi
on-
Programming–
NVPisusedforprov
idi
ngfault
-t
oler
ancei
nsoft
war
e.Inconcept
,theNVPschemei ssi
mil
art
o
t
heN-modularr
edundancyschemeusedtopr
ovi
det
oler
anceagainsthar
dwar
efaul
ts.

TheNVPi sdef inedast hei ndependentgener ationof f


uncti
onall
yequival
entpr ograms,called
versions,from t hesamei ni
tialspecif
icati
on.Independentgenerati
onofpr ogramsmeanst hat
thepr ogrammi ngef f
ortsar ecarriedoutbyN i ndiv
idual
sorgr oupsthatdonoti nteractwith
respectt ot hepr ogrammi ng process.Whenev erpossible,di
ffer
entalgori
thms,t echni
ques,
programmi ngl anguages,env ir
onment s,andtoolsar eusedi neacheff
or t
.Inthistechnique,N
program v ersionsar eexecut edi npar al
l
eloni denticali
nputandt heresult
sar eobt ai
nedby
votingont heout putsfrom t heindivi
dualpr ograms.Theadv antageofNVPi sthatwhena
versionfail
ur eoccur s,noaddi ti
onaltimei srequiredforreconfi
guri
ngthesy stem andr edoing
t
hecomput
ati
on.
ConsideranNVPschemeconsi stsofnpr ogr
amsandav ot
ingmechani sm,V.Asopposedt o
theRB appr oach,al
lnal ternati
veprogramsareusual l
yexecutedsi mult
aneouslyandt heir
result
s are sentto a decision mechanism which sel
ect
st he finalresult
.The decision
mechanism isnormall
yav oterwhentherearemorethantwov er
sions(or,morethankversions,
i
ngener al
),anditisacompar at
orwhenthereareonlytwoversi
ons( kversi
ons).Thesyntacti
c
struct
ureofNVPi sasfol
lows:
seq
par
P1(
ver
sion1)
P2(
ver
sion2)
……
Pn(
ver
sionn)
deci
sionV
I
tisworthwhil
et onotethatt hegoaloftheNVPappr oachistoensurethatmultipl
eversionswil
l
beunli
kelytofailonthesamei nputs.Wit
heachv ersi
onindependent
lydevelopedbyadi f
fer
ent
pr
ogrammi ngteam,desi gnappr oach,et
c.,thegoali
sthattheversi
onswi l
lbedi f
fer
entenough
i
norderthattheywillnotf ai
ltoooftenont hesameinputs.However,
mul t
iver
sionprogrammi ng
i
ssti
llacontroversi
altopic.

Themain diff
erence 
betweent
herecoverybl
ockschemeandtheN-versi
onprogr
ammingist
hat
themodul esar eexecutedsequent
ial
lyinthef ormer
.Therecoveryblockgener
all
yisnot
appl
i
cablet ocri
ti
calsyst
emswher er
eal-
ti
mer esponsei
sofgr
eatconcern.

3.NSel
f–Checki
ngPr
ogr
ammi
ng
NSelf–CheckingProgr
ammi ngi
sacombi nati
ontechniqueofboththeRecov er
yblockandthe
N–v ersi
onProgrammingtechni
ques,whichal socall
sf ort
heacceptancetestexecuti
on.I
tis
per
formedbythesequenti
alandt
heparal
lel execut
ionofv ar
iousv
ersi
onsoft hesoft
ware.
4.ConsensusRecov
eryBl
ocks
InthisTechni
que,
inaddit
iontot heRecoveryBlockandtheN-VersionPr ogr
ammi ngtechni
ques,
thedecisi
onalgor
it
hm techniqueiscombi nedforhandl
i
ngandr ecoveringtheinaccuracyi
nthe
system.Thiscombinati
onofal ltheeff
icientfaul
ttol
erancetechniquesgi v
esamuchmor e
consist
entmethodofFaulttoler
ance.
Adv
ant
agesofFaul
tTol
erance
adv
ant
agesofempl
oyi
ngt
heFaul
tTol
erancet
echni
quesont
othesof
twar
esy
stems,
 Ensur
esasuccessf
ulsy
stem Recov
eryr
out
e
 Achi
evesFunct
ional
Stabi
l
ity
 Hi
gherSy
stem Secur
it
yduet
othecl
ear
lysetr
ecuper
ati
onpl
ans
 Ver
ylessorNoRedundancy
 Al
l
owsPar
all
elpr
ocessi
ngandMul
ti
-t
aski
ng
 CostEf
fect
ive,
ast
hef
aul
tsar
eident
if
iedt
henandt
her
e
 Hi
gherEf
fi
cacyandsy
stem Per
for
mance
 Dat
aaccessi
bil
i
tyi
sper
sist
 Ov
eral
l
,thesof
twar
eisobser
vedt
obeRobust
 Modul
ari
tyhel
psi
nlowerr
eli
abi
l
ityamongstt
hecodeuni
ts
 Doesn’
tlett
hesof
twar
esy
stem t
ocol
l
apse
Concl
usi
on
FaultToler
anceisacr ucialpartofthesof t
warepr ogrammi ngort hesof t
war edev elopment
process,asitismainlyusedt oi denti
fyt heimperfectionsinthesy stem.Af t
ert hebuildis
completedbyincor
porati
ngt heFaultTolerancemethods, i
tishi
ghl
yunl i
kelyforthesof twareto
causeaf ai
lur
eorbreakdown.Hencei ti
sal waysconsider edtoanessent i
alpartint heprocess
ofdev el
opi
nganer r
orfreesof t
waresy stem,especiall
ynow,si ncethetechnol ogicalgrowthis
hugeandnoonewi shest ofacealossi nproducti
vi
tyduet othesof
twarepr ogram errors.

You might also like