Software reliability-growth test and the software reliability-testing platform
Guowei He · 2002
Software users are interested in the MTTF value of software. The rules of /spl lambda/ Pareto diagrams are in general quite different from one software to another. We cannot get an analytical expression to describe the pattern of /spl lambda/ Pareto diagram and be generally suitable to all software. We have constructed a software reliability testing platform (SRTP) which can automatically produce random input points for software following the rule of the f(P). This platform will check the output. If a failure happens, the software will be examined and the defect will be removed, then the reliability of the software will be increased. If the time interval I of software reliability growth test (SRGT) is not quite long, the failure rates of defects appearing within one interval may have no significant differences. If we assume that these /spl lambda//sub i/s are mutually equal to each other we find that the later failure data are closely in compliance with the Duane model. Some authors point out that in some cases they achieve an unreasonable result: the total number of defects N estimated is much less than n (the number of defects appeared). Dr. Song proves that, N has at most n roots and there is only one root which lies between n-1 and /spl infin/ in practical cases. The numerical results of SRGT for a software tested by Dr. Gong and Prof. Zhou are taken as an example. Taking the "zero-failure test" as the verification test of software reliability, the zero-failure test time T needed is T=/spl theta//sub 1//C, where /spl theta//sub 1/ is the minimum acceptable MTTF value, C=-1/ln(1-/spl gamma/), and /spl gamma/ is the s-confidence level.