An Empirical Study on Software Aging Indicators Prediction in Android Mobile

Abstract

The requirements for high reliability, availability, and performance of mobile devices have increased significantly. Android is the most widely used mobile operating system in the world, and it is affected by software aging, resulting in poor responsiveness. This paper investigates the software aging indicators prediction in Android, focusing on aging indicators such as system's free physical memory and application's heap memory. Due to the various user behavior sequences for Android applications and system, we utilize Long Short-Term Memory Neural Network (LSTM NN), which could capture the hidden long-term dependence in a time series to predict these aging indicators. We analyze the prediction results with traditional evaluation metrics like MAPE/MSE for evaluating the whole prediction performance, and with our proposed evaluation metrics TA, FA, SVA for evaluating the trend, fluctuation, and small variation of aging indicators respectively. The results show that LSTM NN has superior performance compared with other prediction methods in the history of software aging researches. Based on the results, proactive management techniques like software rejuvenation could be scheduled by predicting the proper moment to alleviate software aging effects and increase the availability of Android mobile.