Survey of Dynamic Instrumentation of Operating Systems

Operating systems and modern software systems get more complex and more difficult to understand every day. Instrumentation of a software system is a technique for inserting extra code into an application to observer its behavior[6]. It provides insights to a running system, and helps to understand even complex systems. The information provided by an instrumentation tool can be used do debug a running system, to improve the performance of a system by identifying bottlenecks, or simply help to understand the control flow of an application or system. In [6], the authors explain how they used their instrumentation tool to identify a stock ticker applet that was responsible for poor performance on a multi user system with more the 170 users. Just by killing the applet the system idle was increased by 15 percent. Instrumentation can be divided in two major categories: static instrumentation and dynamic instrumentation. Static instrumentation refers to instrumentation techniques that provide information by inserting instrumentation code before the program runs. The additional code could be inserted by the developer in the source code of the program, by the compiler, or by linker. The instrumentation code is always present in the running program, and cannot be changed without stopping or even recompiling the program. The instrumentation statements, however, are usually guarded by a flag to dynamically enable or disable the instrumentation.