A low-overhead design of context-sensitive profiler for Android applications

The computing of applications on mobile devices suffers tight constraints on computation and energy resources. Thus, it´s important for embedded applications running on these resource-constrained devices are able to execute efficiently. The Android system uses dalvik virtual machine to execute Java applications, and the virtual machine causes bottlenecks of the application hard to be identified directly by existed profiling tools, such as Oprofile. This paper proposes a design of a low-overhead profiling tool embedded in the dalvik virtual machine to discover the bottlenecks of Android applications with hardware performance counters. The experimental results show that the bottlenecks of the Android application can be identified efficiently by the proposed profiler.