Research On Macroscopic Topological Characteristics Of Software Networks And Verification In Large-scale Hydropower Software

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. With the development of the speculation and the complexity of software systems, the interaction of the constructive parts makes the software system as a whole appear a new feature which contains in the construction and metabolism of the system. There are various forms of integration in different levels, different sizes, and different organizing types. Measurements always play a key role in various systems, which is helpful to understand the system in full view and improve system performance by analyzing and improving. With development of software technology and application requirement, new requirements of method of metric and analyzing are emerging. Traditional method only aim at complexity of code or Structure of Structured Program, there is still no more effectively method for the quantitative measurement of software. How to master and control the increasing complexity of software system and ensure its dependability from various respects is the key problem for software metric to resolve and is also a great challenge for software engineering.With deeply development of research on theory of network, it is applied by scientists to research on complex system, i.e. complex network. Complex network is a network which is composed of many units and there are closely relationship and interaction between the units. Because of multi-granularity of software(function?class?source code?program library? component and etc), association and interaction of which can be abstracted to a kind of network topology, and then can be researched as a model of structure characteristics. In recent years, applying theory of flex network topology to software metric has been proposed and some research has been carried out. Hence, there comes the concept of software network, it is helpful to understand basic properties and law of software structure in full view and reflect the influence of structural characteristic on system quality. These theories of macroscopic topological characteristics are being perfected constantly in recent years.First, the background and current situation of software network are introduced, and also the main research content and significance and in this paper. It contains theory of complex networks and metric method, current situation of applying theory of complex networks to researches on large-scale software structure and the extracting process from software system to software network. Large hydropower software Wemux and selecting rules of software samples are introduced.Second, multi-granularity i.e. function?class and package software network is introduced by figures and illustrations. A new method is suggested to metric and analyse the different granularity of software network characteristics to find the rules which are contained in software architecture of some software including Wemux. The conception software network topology?degree?degree-distribution?clustering coefficients?betweenness?average shortest path?and closeness are presented, and then analyze the software network, it shows that there is a significant power-law correlation between distribution of degree distribution in the undirected graph and out-in degree distribution in the directed graph. It reflects the modular design idea. The node whose degree is low has high aggregation effectiveness, which is inclined to cluster closely to others. There is high aggregation effectiveness in multi-granularity software network. The node which has large betweenness often has low degree, it shows that controlling complexity of key modules was considered by designers and developers of Wemux. An algorithm of betweenness is proposed and results show that calculation time is acceptable on package and class network. Analysis of closeness shows that degree of central node often has low degree, the value of closeness dosen't increase obviously along with degree increases. Wemux is taken as an empirical to verify validity of multi-granularity metric method, and then characteristics and guiding significance are illustrated.Third, Class is the most important data structure and also the basic node on software network. Hierarchy and community?interweavement?assortative coefficient?improved structural holes and etc are proposed to metric and analyze the structural version evolution and cooperate modularization. The algorithm of circular layout is proposed for clear display. The results show that:There is an significant correlation between number of nodes and path, the dependency to some key node increase in a certain extent along with the scale of system increase; Interweavement firstly increase and then decrease along with the scale of system increase; In the process of scale increasing, edges in the community is firstly more rapid than those not in community and then close, coupling between modules becomes steady. There is not a significant correlation between average shortest path and assortative coefficient, but average shortest path decrease along with assortative coefficient increase; Improved structural holes can better reflect bridge role of node on software network, there is a significant power-law correlation between the degree and structural holes. Wemux is taken as an empirical to analyze, metric significance is illustrated with class module.Fourth, wave effectiveness of node is proposed, based on which, node loads multiple usages multiple reliance are proposed. Node load is firstly introduced into software network to predict fault-proneness, which is widely used in other fields. Valve class is taken as an example to illustrate how to compute the node load. The results that:Node load considering effectiveness of the class node on other nodes. How to compute fractal feature of node load distribution is illustrated. The results show that:There is not a significant correlation between number of node and fractal feature of node load distribution, but there are different threshold ranges according to the kind of software. Wemux is taken as an empirical to show that node load can reflect some important characteristics, multiple usages and multiple reliance can reflect complexity and structure defect of class effectively.Eventually, a platform of metric and analysis is implemented on software network. The overall architecture is introduced, storage and computation is implemented respectively. Users needn't spend much time to get the results. Structure design of main tables is illustrated, Macro and micro characteristics are stored respectively. Solutions to key problem of modules are demonstrated in detail.