| As one of the effective means of expressing human intelligence, software is different from the general industry product whose quality can be constrained by strict testing standards. So, it is hardly to control and ensure the quality of a software product. When software systems play more and more important role in our life, there is a leap of the scale and complexity of them with the increasing function demand. As for the software developers, to grasp the whole internal structure consisted of quantities of component elements and their relationships between each other is too difficult to control the developing course, which lead to high cost for maintenance cost. How to master and control the increasing complexity of software system and ensure its dependability is the key problem for software metric to resolve and is also a great challenge for software engineering.Donte Chalf, a physiologist of Northwestern University in USA, believes that human cerebral structure is a complex network, which present strong small-world characteristic, and form effective connection. As product of human intelligence, software network is a fractal of cerebral structure, and its internal structure also present significant characteristics of complex network. The characters and regularities of internal structure can be understood from view of the whole system by describing the complex composition of software with network model. In this dissertation, a directed and weighted network model is established by abstracting the software internal structure. Based on this model, the structural characteristics which have effect on software dependability are discussed and the further research on dependable properties is carried out.Firstly, after analyzing the lacks of traditional software metrics applied for large-scale systems, a multi-dimension method system on software metric is presented according to the structural characteristics’effect on the quality of software. In this system, metrics for the structural complexity, stability and reliability of a software system are discussed separately in three dimensions which are macro-topology, micro-composition and foundation structure. According to the analysis on structural characteristics in these three dimensions, the metric parameters and methods for quality of software are determined.And then, the metric methodology of dependability of software system based on three dimensions of software structure is discussed in detail in the following parts of this dissertation. On the dimension of micro-topology, the statistics of main eigenvalues in a large number of sample software systems and their distribution test by K-S method are analyzed in the first step. According to the eigenvalues’effect on the structural complexity, executing efficiency and orderliness, a metric method based on deviation of eigenvalues is proposed, by which the quality characteristic relative scale of software system can be measured. On the other hand, according to the mode and characteristics of the system composition, a metric method based on composition characteristics is proposed. The definitions of radiation degree and radio of radiation cycle play a role in metrics for complexity of system composition and structural defects. The result of this metrics will provide guidance for software developer during testing or maintaining the system.Secondly, the structural stability of software system is analyzed and measured on the dimension of micro-composition. Because of the reusability of design patterns in object-oriented software system, the structure of design patterns is studied as typical micro structure which have effect on the whole software network. So a metric method for structural stability based on design patterns is presented in this dissertation. Using Lyapunov method, the stability of local structure abstracted from design patterns’composition is measured, and the result shows that all of the patterns have high structural stability. According to the application of23design patterns in sample software systems, coverage of design patterns is defined as the main metric parameter for the stability of whole structure, which reflects the effect on the stability of software system by design patterns.Thirdly, the definition of software-core structure, which presents the foundation structure of software network, is proposed. Because of the critical effect on the reliability of software system, the reliability of software-core structure deserve for research. The distribution of degree in software-core structure is analyzed in order to determine the reliability of it according to the effect of inheritance on dependability of system. During the course of extracting software-core structure from software network, the distribution of inheritance trees and the inheritance depth of each node in software network can be gathered. According to the relationship between inheritance depth and degree, inheritance dependency of class node is defined to detect the structural defects in design which have deep inheritance and large in-degree at the meantime.Eventually, based on the multi-dimension metric system discussed in this dissertation, a platform of metric and analysis on software dependability is designed and implemented. Using this platform, the internal structure of open source software systems can be analyzed directly and the dependability characteristics of software are measured on three dimensions. Further more, the platform provides visual internal structure of software. It is helpful to master the structural complexity and its effect on system quality, which presents guiding significance to development or maintenance of software system. |
PDF Full Text Request