Theory And Method Of Component Based Software Reliability Analysis

With the development of computer software technology, the traditional Software Reliability Growth Model (SRGM) which is based on black box can not meet the requirements of complex software reliability analysis presently. Since it is only supported by collecting system failure data in the late stage of software life circle and it ignores software architecture information entirely. On the other hand, the Architecture Based Software Reliability Model (ABSRM) has gotten more attention within the last decades because it can combine component reliabilities with the architecture information effectively and can assess system reliability more rationally and accurately. For the characteristics mentioned above, the ABSRM method will play an important role in the future of software reliability engineering field for it is very suitable in handling the reliability problems of large-scale component-based software. In this dissertation, the ABSRM is also chosen as the main object of studies, focusing on the problems of analyzing the reliability effectively in complex component-based software system. Specifically, these problems involve three aspects, namely:1) complex software reliability modeling;2) software reliability assessment in the early phase;3) parameter analysis of the software reliability model. The main works are summarized as follows:(1) As the beginning of studies, this dissertation elaborates the significance of the studies on software reliability modeling, reliability assessment in the early phase and parameter analysis of reliability model, and points out the relations among the three aspects and the main object of studies. Moreover, the surveys of the studies on the three aspects are given and the existing problems are pointed out simultaneously.(2) In this dissertation, the formal description method for non-Markov control transitions among components is studied. For the deficiency of traditional Architecture Description Language (ADL) in characterizing the component interaction modes, a supplement is given and the corresponding mapping algorithm is proposed also. Based on the above, a new software reliability model called CIM (Component Interaction Mode) is proposed which can be used for considering the cases of component interactions with non-failure independent and non-random control transition. Through this model, the component interactions can be described with a new formal description method based on abstract algebra theory, the additive nodes and the transitions probabilities among them can be constructed and calculated. The new architecture diagram derived from the model mentioned above can satisfy the Markov property and its control transition probabilities matrix is identified certainly. At last, the reliability analysis process for a practice software system is given to illustrate the effectiveness of CIM.(3) In this dissertation, the quantitative relationships among system reliability and software architecture styles are analyzed and the Basic Architecture Reliability Paradigm (BARP) is constructed. Besides that, a reliability evaluation method of complex software is proposed. Firstly, it establishes the formalized classification for the different software architecture styles with their functional definitions in SA (Software Architecture) theory and the formalism descriptions of their components and the component interactions. Then based on the component failure independence verification, the failure dependent relationships among components are judged, because of which we can obtain the system states and the transitions among them from the information of components. Finally, the software architecture is modeled as a state transition model, which is based on the Discrete Time Markov Chain (DTMC) theory, so the reliability problem can be solved as the N steps transition probability problem involving the parameters such as the component reliabilities, the matrix of control transition probabilities and the component execute frequencies. With the method proposed, we can get the certain math expressions of reliability paradigms for different software architecture styles, and each of them can be mapped into an independent system state following the conversion mechanism already constructed, and the corresponding parameters can be calculated with certain formulas. Moreover, for illustrating the influence in reliability evaluation of the operational profile, the frequency of component execution is taken as a parameter. By comparing the results of the proposed approach with some other algorithms through evaluating the reliability of a practice software system, the rationality and accuracy of reliability analysis are proven promotion.(4) In this dissertation, we study and establish a model conversion method for software reliability analysis based on software design models UML. UML models are the most important products in the early stage of software development, and they provide a wealth of valuable information for the different aspects of the whole system. For the purpose of simplifying the complexity of reliability analysis, reducing the workload of software reliability assessment in the early phase, and making the model can deal with multi-styles situation, an approach is proposed in this dissertation that can predict complex component-based software reliability with the UML models directly. This method is based on the existing relevant researches of software reliability modeling and analyzing, it utilizes three types of UML diagrams:use case, sequence and component diagrams, extends them by annotating the necessary attributes related with reliability evaluation for making it have the ability to descript the multi-styles architecture. Then the extended diagrams are transformed into the reliability analysis model DTMC by constructing an intermediate model called System State Transition Diagram (SSTD) and establishing the corresponding algorithms. At last, the system reliability can be evaluated with this method mentioned above.(5) In this dissertation, we make parametric analysis for complex software reliability model, because of the inaccuracy of parameter estimation and the demands of key parameters judgment method. For the former problem, a new simulation analysis method is proposed in this dissertation, it combines the advantages of SRGM in characterizing component reliability growth (using the time-dependent failure density function μ(n, t) to characterize the failure behavior of components) with the advantages of ABSRM in modeling software architecture (using DTMC to represent system composition). Through this method, the influences of system reliability from component reliability growth caused by the error recovery can be analyzed quantitatively in software test phase. In this simulation analysis method, the failure behavior of components and the system are model to Non-Homogeneous Continuous Time Markov Chain (NHCTMC), besides that, the processes of component failure, system failure and error recovery are simulated. With the simulation method proposed, the five key software reliability analysis elements; which are the component reliability, the error detection rate, the system architecture information and the error recovery rate, can be associated as a whole entity. It provides a new idea for large complex component-based software reliability analysis in software testing phase. For the latter problem, a key parameter judgment method of software reliability prediction with the moment estimation theory is proposed which involves the statistical characteristics of every parameter. It considers the component reliability, the control transition probability matrix and the system reliability as random variables. Moreover, the quantitative relationships among the system reliability prediction result and the moment estimations of the parameters are explicitly given. This judgment method involves the operation profile and all the other relevant parameters in ABSRM, and it takes into account the impacts on system reliability from them. Compared with the traditional point estimation method, the approach proposed can provide more information of system reliability, what is of great significance for early software reliability prediction, system reliability control and reliability tracking in the software evolvement.