Study On The Efefct Analysis Of Testing Methods On Software Reliability

With the development of technology, software has been deep into all aspects ofhuman life. Software is of great benefits to human life and it makes people live anefficient and convenient life, however, more and more system failures occur due tosoftware quality. Software reliability is a key element among all factors relating tosoftware quality. How to improve the software reliability, which is just like the issuethat how to improve software productivity, is an important problem which need to bepaid attention to during the whole software life, and it is also an important problemwhich we must try our best to find solution for. Software reliability computing relieson software reliability model and failure data. The selection of software reliabilitymodel is relevant to software execution environment, while failure data relates to theapplied software testing strategy. In traditional software reliability computingapproach, different failure data will be derived when different testing strategies areapplied, as a result, software reliabilities obtained from different strategies are not thesame. In this case, the computed result only represents the software reliability undersome specific environment, which is static but not the software reliability that candescribe the actual information of software.We propose a new method for software reliability computing and our methodmakes use of program invariants. Program invariants denote properties which keepunchanged during program execution, and they depict the dynamic behaviors of aprogram and can represent the whole behavior of software, in this way, the softwarereliability computed by our method is more exact. We execute software system with aset of test cases to dynamically extract program invariants, then we collect failure data,and at last, we compute software reliability by using Nelson model. Different teststrategies may generate different invariants and different failure data, but softwarebehaviors derived from these test strategies must be the same because of thecorrelation between invariants for one given test strategy and program behaviors canbe synthesized from all of these invariants. Our experimental results show that software reliabilities computed through three different testing strategies, that is,random testing, branch-coverage testing and block-coverage testing, are almost thesame.The contributions of this paper are listed below.(1) Invariants based software reliability computing approach can capture softwarereliability at runtime, so the computed result is more exact and is more close to theactual runtime situations of software.(2) Testing strategies have little influence on invariant based software reliabilitycomputing approach, so selection of an appropriate testing strategy for computingreliability becomes an easy task.(3) Failure data extracted from invariants is helpful to evaluate software quality,monitor and predict software execution situation, also it provides a new way for us toimprove software quality.