Predicting for I/O stack optimizations on cyber-physical systems

Yangmei Zhang, Fanfan Shen, Mengquan Li, Chao Wu

Abstract: I/O stack has been identified as the key component to system performance on cyber–physical systems, e.g., mobile devices. A lot of works have been proposed on I/O stack optimization. The state-of-the-art operating systems need to perform many system management operations to achieve better performance. However, the additional managements activities of operating system could greatly affect user experience if these activities are triggered at improper times. This paper presents a novel prediction framework so that system management activities can be performed at IO idle time to avoid impact on user experience. First, it collects mobile device using information. With the collected information, the proposed framework predicted IO idle time which can be used to handle system activities in a cost-free manner. Evaluation results show that the proposed framework predicts I/O state with accuracy of 78.62% on average while introducing negligible overhead.