Poster: Towards reducing smartphone application delay through read/write isolation

Abstract

A recent analysis[3] indicates that most user interactions with smartphones are short. Specifically, 80% of the apps are used for less than two minutes. With such brief interactions, apps should be rapid and responsive. However, the same study reports that many apps incur significant delays during launch and run-time. This work addresses two key research questions towards achieving rapid app response. (1) How does disk I/O performance affect smartphone app response time? (2) How can we improve app performance with I/O optimization techniques? We address the questions via following contributions. First, through a large-scale measurement study based on the data collected within 130 days from 1009 Android devices using an app[2] we developed, we find that Android devices spend a significant portion of their CPU active time (up to 58%) waiting for storage I/Os to complete, also known as iowait (Figure 1). This negatively affects the smartphone’s overall app performance, and results in slow response time. Further investigation reveals that a read experiences up to 626% slowdown when blocked by a concurrent write. Additionally, the results indicate significant asymmetry in the slowdown of one I/O type due to another. Finally, we study the speedup of concurrent I/Os, and the results suggest that reads benefit more from concurrency.