Better Performance Through Thread-Local Emulation

Mobile platforms are shifting away from managed code and toward native code. For example, the most recent versions of Android compile Dalvik bytecodes to native code at install time, and apps frequently use third-party native libraries. The trend toward native code on mobile platforms calls us to develop new ways of building dynamic taint-tracking tools, such as TaintDroid, that achieve good performance. In this paper, we argue that the key to good performance is to track only when necessary, e.g., when an app handles sensitive data. We argue that thread-local emulation is a feature that captures this goal. In this paper, we discuss the motivation for thread-local emulation, the software and hardware techniques that may be used to implement it, results from preliminary work, and the many challenges that remain.