Profiling the Responsiveness of Android Applications via Automated Resource Amplification

The responsiveness of the GUI in an Android application is an important component of the user experience. Android guidelines recommend that potentially-expensive operations should not be performed in the GUI thread, but rather in separate threads. The responsiveness of existing code can be improved by introducing such asynchronous processing, either manually or automatically. One simple view is that all potentially-expensive operations should be removed from the GUI thread. We demonstrate that this view is too simplistic, because run-time cost under reasonable conditions may often be below the threshold for poor responsiveness. We propose a profiling approach to characterize response times as a function of the size of a potentially-expensive resource (e.g., shared preferences store, bitmap, or Silt database). By manipulating and "amplifying" such resources automatically, we can obtain a responsiveness profile for each GUI-related callback. The profiling is based on a static analysis to generate tests that trigger expensive operations, followed by a dynamic analysis of amplified test execution. Based on our evaluation, we conclude that many operations can be safely left in the GUI thread. These results highlight the importance of choosing carefully - based on profiling information - the operations that should be removed from the GUI thread, in order to avoid unnecessary code complexity.