Research Of Software Aging And Rejuvenation In Android

Software aging refers to the decline of software performance due to the accumulation of error during the process of long-term continuous operations.Software rejuvenation is a cost-effective method to against the effects of software aging.Similar to the traditional hardware aging,software aging also has the characteristics of randomness.But its frequency of occurrence is higher,while its detection and analysis are more complex,and the damage is greater.It is one of the emerging research hotspots in the field of software reliability engineering in recent years.The share of Android phones in the market accounts for the majority.A common phenomenon is that as the Android phone usage time increases,the UI response time of the mobile phone increases,the performance gradually decreases,and the user experience is relatively worse.This phenomenon may be caused by the software aging in Android.Studying software aging in Android can improve the performance of Android,improve system reliability and enhance user experience.In this thesis,we build experimental schemes and collects memory-related data for modeling and analysis,to verify the existence and characteristics of software aging and the effect of traditional warm rejuvenation strategy.Then we use the random forest method to judge the aging state.Finally,we combine the usage behavior in mobile intelligent terminal to build a software rejuvenation model,and propose the dependability-related metrics that comprehensively considers the usage behavior and user experience.The main work is as follows:1)Through the stress test experiment and modeling analysis,the existence of Android software aging is verified,and it is found that aging is generally irreversible,and warm rejuvenation(application restart)has limited rejuvenation effect on software aging in Android.A lot of research has shown that the memory leak is a major cause of software aging,and because of the complexity of Android memory management mechanism,we build stress experiment through the black box test method to observe the changes of the available memory of Android,and gradually verify some relevant assumptions about the software aging in Android.Then we use the discrete time Markov model to model and analyze the collected data.2)By using the random forest algorithm to judge various states in the process of software aging in Android,we provide better guidance for the software rejuvenation in Android.Since different workloads on Android have different aging phenomenon,which may influence the validation of system state.Based on these,we carried out experiments to simulate user behavior on android,and collect relevant data,then use random forest algorithm to judge the state of the system.Experiments show that the method has a lower recognition error rate.Finally,we construct a corresponding rejuvenation strategy based on the recognition model,which could provide advice on the development of a better software rejuvenation strategy in the future.3)We combined the characteristics of mobile phone usage behavior to construct the software rejuvenation model,and proposed two quantitative analysis metrics: Fluency Availability and User Experience Degree.The mobile intelligent terminal includes mobile phones have the characteristics of fragmentation and periodicity of usage behavior.Therefore,we first model the usage behavior and aging processes separately,then combine them with a stochastic Petri net and transform them into a continuous-time Markov chain;on this basis,by considering the behavior patterns and aging process to construct a rejuvenation strategy to achieve the best user experience and least user interference,such as restarting the device when it is in the sleep mode and before it enters the aged state(the UI response is slow);while the traditional dependability metrics do not including the quantitative analysis metrics reflecting the user experience,based on the model we constructed we propose two new metrics: Fluent availability(the probability that users can smoothly use the mobile phone)and User Experience Degree(The overall experience of the user experience of the system);finally,the effectiveness of the proposed model and metrics is verified by numerical experimental analysis.