基于软件网络的软件系统演化规律验证和模拟  

Validation and simulation of software system evolution rules us- ing software networks

作　　者：顾庆[1] 陈道蓄[1] 

机构地区：[1]南京大学计算机科学与技术系软件新技术国家重点实验室,南京210023

出　　处：《中国科学：信息科学》2014年第1期20-36,共17页Scientia Sinica(Informationis)

基　　金：国家重点基础研究发展计划(973)(批准号:2009CB320705);国家自然科学基金(批准号:61373012;91218302;61021062);国家高技术研究发展计划(863)(批准号:2006AA01Z177)资助项目

摘　　要：演化是计算机软件的固有特性.了解演化规律,可以改进软件演化过程,提高软件质量,降低维护成本.本文定义类依赖图,将软件系统描述为软件网络,基于网络度量验证软件演化规律,并讨论软件演化建模的可行性.首先针对Lehman的8项演化定律,设计3个方面的网络度量:网络规模、网络质量、以及结构控制.然后基于四个开源软件系统构建软件网络,统计网络度量;实证研究表明:支持Lehman的4项演化定律,否定另外3项,即复杂度提升、持续增长、以及质量折旧定律.最后分别采用E-R模型和模块依附模型模拟软件系统的演化过程,所生成的随机网络同软件网络的比较结果说明,软件演化存在客观规律,能够通过建模来重现甚至优化.Evolution is an intrinsic property of computer software. By understanding the dynamics of evolution,the process of software evolution can be improved,so that the quality of software can be improved,while the maintenance cost can be saved. By defining the class dependence graph,this paper transforms software systems into software networks,validates software evolution laws using network measures,and discusses the feasibility of modeling software evolution. Firstly,according to the eight evolution laws provided by Lehman,network measures are designed and grouped into three aspects： network size,network quality,and structural control. Secondly,based on the four selected open-source software systems,software networks are built and network measures are counted. The experimental study proves that,four evolution laws of Lehman are supported,while three of the rest,i.e. ＂increasing complexity＂,＂continuing growth＂ and ＂declining quality＂ have contradictory cases. Finally,both the E-R model and the modular attachment model are used to simulate the evolution process of the software systems. The results of comparison among generated random networks and software networks prove that intrinsic laws exist in software evolution,and by prudent modeling,software evolution can be replicated and even optimized.

关 键 词：软件演化 复杂网络 软件演化定律 类依赖图 随机网络 

分 类 号：TP311.52[自动化与计算机技术—计算机软件与理论]