A
A
A
AStudy
S tudy
Study
Studyon
o n
on
onSoftware
S oftware
Software
SoftwareReliability
R eliability
Reliability
ReliabilityPrediction
P rediction
Prediction
Predictionbased
b ased
based
basedon
o n
on
onTriple
T riple
Triple
TripleExponential
E xponential
Exponential
ExponentialSmoothing
S moothing
Smoothing
Smoothing
Method
Method
Method
Method(WIP)
( WIP)
(WIP)
(WIP)
Jinyong
Jinyong
Jinyong
JinyongWang
Wa ng
Wang
Wang,
,
,
,Zhibo
Z hibo
Zhibo
ZhiboWu,
Wu ,
Wu,
Wu,Yanjun
Y anjun
Yanjun
YanjunS
S
S
Shu
h u
hu
hu,
,
,
,Zhan
Z han
Zhan
ZhanZhang
Z hang
Zhang
Zhangand
a nd
and
andLixing
L ixing
Lixing
LixingXue
X ue
Xue
Xue
Department
Department
Department
Departmentof
o f
of
ofComputer
C omputer
Computer
ComputerScience
S cience
Science
Scienceand
a nd
and
andTechnology
T echnology
Technology
Technology
Harbin
Harbin
Harbin
HarbinInstitute
I nstitute
Institute
Instituteof
o f
of
ofTechnology
T echnology
Technology
Technology
Harbin
Harbin
Harbin
Harbin150001,
1 50001,
150001,
150001,China
C hina
China
China
wangjinyong818@163.com
wangjinyong818@163.com
wangjinyong818@163.com
wangjinyong818@163.com,
,
,
,{wzb,syj
{ wzb,syj
{wzb,syj
{wzb,syj,zz,xuelixing
, zz,xuelixing
,zz,xuelixing
,zz,xuelixing@ftcl.hit.edu.cn}
@ ftcl.hit.edu.cn}
@ftcl.hit.edu.cn}
@ftcl.hit.edu.cn}
Abstract
Abstract
Abstract
Abstract
Inthesoftwaretestingprocess,theeffective
developmentofsoftwarereliabilitymodelscanbe
quantifiedtoassessthesoftwarereliability.Most
ofthecurrentsoftwarereliabilitymodels
developedaretheparametricmodels,and
assumingthatthesoftwarefaultdetectionprocess
isMarkovornon-homogeneousPoissonprocess
(NHPP).However,duetothecomplexityofthe
softwaretestingprocess,assumptionsofbuilding
themodelscannotsatisfytheactualsoftware
testingsituation,andthoseparametricmodelsset
upbyassumptionscannotaccuratelypredictthe
softwarefailureprocess.Inthiscase,wepropose
anon-parametricmethodofthesoftware
reliabilitypredictionbasedonthetriple
exponentialsmoothing.Wecomparetheproposed
non-parametricmethodwiththedouble
exponentialsmoothingandothersoftware
reliabilitymodel.Theexperimentalresultsshow
thatusingtheproposednon-parametricmethod
caneffectivelyandaccuratelypredictthenumber
ofsoftwarefailure.Furthermore,itdoesnotneed
alotofhistoricalfaultdataandcalculations,and
canbeeasilymadetoolusedintheactual
softwaretest.
Keywords:
Keywords:
Keywords:
Keywords:Softwarereliability,tripleexponential
smoothing,reliabilityprediction
1.
1.
1.
1.INTRODUCTION
I NTRODUCTION
INTRODUCTION
INTRODUCTION
Withtherapiddevelopmentofcomputer
technology,alargenumberofcomputersapplyin
thehumansociety,andbringalotofconvenience
forpeopleliveandwork.Itishopedthatthe
softwarerunningonthecomputercanbethe
failure-freerunningforever.Therefore,software
reliabilityhasbeentheproblemofparticular
concern.But,infact,owingtothecurrentsizeand
complexityofsoftwaredevelopmentarefarmore
thanbefore,inevitablyfaultsexistinthesoftware
afterthereleaseofsoftware.Inordertoreducethe
probabilityofsoftwarefailurerunningonthe
computer,beforethesoftwareisreleased,itwill
beconductedmanysoftwarereliabilitytests.In
thesoftwaretestingprocess,usingthesoftware
reliabilitymodeltoassessthereliabilityofthe
softwareisagoodmethodforquantitative
evaluationofsoftwarereliability.
Overthepastfourdecades,researchershave
developedanumberofparametricsoftware
reliabilitymodels,suchas,Musaetal.[1],Pham
[2]andLyu[3];tosomeextent,thesemodelscan
effectively assesssoftwarereliability,suchas
providingsomeusefulinformationonwhento
terminatetestandreleaseanewversionof
software,etc..However,thetraditionalparametric
softwarereliabilitymodelsgenerallyassumedthat
thesoftwarefaultdetectionprocesswasMarkov
(Markov)ornon-homogeneousPoissonprocess
(NHPP),andsubjecttocertainprobability
distribution.Forexample,GoelandOkumoto[4]
assumedthatthefaultdetectionprocesswasa
non-homogeneousPoissonprocess,andinthe
processofthesoftwarefaultdetection,thefault
intensitywasproportionaltothenumberof
remainingfaults,andthefaultdetectionratewas
constant.Theyproposedwell-knownG-Omodel.