Research On Architecture-Based Software Reliability Demonstration Testing Method

The technique of software reliability testing and evaluating is one of the important methods to safeguard the reliability of software systems. It is of great significance in the key areas of aerospace, transportation safety. In order to adapt to the multi-tasking and multi-modular trend of the modern high-dependability software, this article establishes a framwork of software reliability testing and evaluating method basing on the information of software architecture, including software system reliability demonstraion target establishment, software system reliability design target establishment, allocation of software system reliability development target, software modular-level reliability growth testing and software system-level reliability demonstration testing. This method not only combines the software reliability growth testing with the reliability demonstration testing, but also fully unifies the software reliability evaluation process, the software development process and the software structure attributes. On the basis of ensurence of reliability requirements, this method solves the key issue in reliability growth and demonstration testing for high-reliability software to some extent that it requires too excessive test cases, and the test duration is too long.This paper firstly describes the idea of software reliability demonstration testing, and then draws a conclusion that the statistical software testing and reliability evaluation are essentially the same after comparing of them. At the same time, we analyze the limitations of the existing software reliability evaluation methods. After making some improvement and innovation of some existing software reliability evaluation methods. we propose our own architecture-based software reliability evaluation framework. The characteristics of the evaluation framework is to organic the software reliability growth testing and demonstration testing as a whole, and to implement effective reliability tracking and control from the software module level.On the research of test resource allocation method in software reliability growth testing, we do a full analysis of previous work, and then propose a dynamic resource allocation method of reliability testing for multi-modules software with Markov transfer of control. The method is particularly used for the Markov model of multi-module software system to achieve rapid growth of system reliability through targeted module-level testing based on their various improving effect on system reliability. And taking into account the dynamic changes of test difficulty caused by each module’s complexity, we dynamic adjust test resources during testing phases to ensure that the test resource allocation is reasonable.On the research of reliability demonstration testing methods, we use the Bayes small sample statistical method to reduce the test cases by making full use of existing software module-level failure data as priori information. Considering the issue of how to take advantage of the system architecture information and module-level prior knowledge, in order to obtain the optimal Bayes priori super-parameters in risk allowable range, we presents the Bayes priori parameters derivation method based on the minimum-risk judgment standard, which will provided strong support for the reliability demonstration testing.This paper proposes the architecture-based software reliability demonstration method after analyzing and summarizing of previous studies, and provides solutions to software reliability target conversion, optimal allocation of resources in software reliability growth testing, as well as software reliability demonstration testing method. Numerical simulations show that the proposed method can effectively reduce the test cost, and is more conducive to software reliability evaluation of engineering practice. Considering the facts that the size and the complexity of the safety-critical software is sharply incresing, and the software reliability testing&evaluating study is relatively slow and lag behind that of hardware, the presentation of above methods has a certain theoretical and practical significance.