GAN-ASD: Precise Software Aging State Detection for Android System Based on BEGAN Model and State Clustering

Abstract

Software applications may become no response or stop running due to performance degradation, system crashes, or program cumulative failures, after long-term execution in Android system. These phenomena have been validated to be common in mobile systems which is caused by software aging. To handle the software aging dilemma, software rejuvenation is an efficient way. In order to make rejuvenation more efficient, identifying the aging state of Android system precisely is the key point. In this paper, we propose a novel Android system aging state detection method based on the Boundary Equilibrium Generative Adversarial Network (BEGAN) and state clustering technology, which is named as GAN-ASD. The method has three phases: Firstly, Interpolation Clipping Processing is used to processes the time series dataset which is constituted by the sample of Android Aging Indicators. Secondly, according to the time series dataset, BEGAN based generation method will fit the user’s usage habits and generate the dataset which has software aging characteristics. At last, we use the generative dataset to train a K-Means clustering model. With the trained model, we can precisely determine whether the current Android system enters into the aging state or remains in the normal state. In order to validate the effectiveness of the GAN-ASD, we use two evaluation criterion in our comparison experiment. One is rejuvenation coefficient (RC) which evaluates the user experience and the other one is rejuvenation frequency (RF) which evaluates the rejuvenation cost. The results show that our method performs better than the fixed-interval rejuvenation and random rejuvenation operations.