Research On Software Behavior Dependence Problem Oriented To Dynamic Evolution

Large-scale and complex function is one of the characteristics of modern software products. Legacy systems leave a huge fortune to the market and the developers, but it can not satisfy people’s changing production business needs, especially for the SaaS service that is open, dynamic, changeable and uncontrollable, which is provided by modern-mainstream-technology-oriented cloud computing and highlights the importance and urgency of software dynamic evolution.Software dynamic evolution theory and technology research bring about not only challenges, but also opportunities to academia and industry. At present, there are two "challenges" in software dynamic evolution theory:the first one is behavioral correlation, before the implementation of software evolution, the scope of evolution must be determined so as to try to control the spread scope causing by the implementation of evolution and reduce the cost of evolution; the second one is about consistency, which must be ensured before and after the evolution of software. In order to obtain the prior knowledge of the software dynamic evolution, the HMM-based approach has been proposed, studying the correlation of behavior transfer at runtime. The observable set of states are parameters, context, actions, status, interfaces and port to build behavioral correlation analysis algorithm, which are all the software behavior attributes. The HMM mathematical model is trained by observable sequence, then, the hidden software system behavior state series is calculated according to the model and the observable sequence, finally, the state of software behavior is represented by the state vector of software components, and then, the correlation between the components is analyzed through the migration of components state vector that is caused by the migration of software system states. The concrete steps are as follows:(1) The component behavior correlation analysis algorithm is constructed on the premise of assumption;(2) The observation index of software behavior related attributes are set: parameters, context, actions, status, interfaces and ports;(3) The state transition matrix from the moment T to T+1 is calculated via the observation sequence of behavior attributes;(4) The confusion matrix is calculated by the HMM optimal model trained by Maximum Likelihood Estimation, then, the probability relationship between software system state transition and behavioral properties is returned;(5) The relevant components set is analyzed by establishing the corresponding relation between system state and component state vector function;(6) The detailed simulation experiment is designed to verify the feasibility and effectiveness of this approach.According to the simulation result, it shows that the approach is feasible and effective.