Performance Evaluation of Dynamic Binary Instrumentation Frameworks

Code analysis, static or dynamic, is a primary mean for improving correctness and performance of software applications. Dynamic binary analysis (DBA) refers the methods to analyse runtime behaviour of binary code. Nowadays, DBA tools are implemented using dynamic binary instrumentation (DBI) frameworks, which can add arbitrary code into the execution flow of the binary. Unfortunately, a DBA tool increases the execution time of the analysed binary dramatically, as extra code is being executed. Experiments got figures with increments of 26x. Therefore, it is crucial for DBA tool construction to know exact figures about such penalties and their roots. Hence, we conduct a performance evaluation of leading DBI frameworks, namely Pin, Valgrind, and DynamoRIO, for which we have built a benchmark selecting a bunch of representative tools. The evaluation that we procure here provides guidance to choose the best DBI framework suited for different needs. Moreover, the benchmark by itself is a tool ready to be eventually used for performance evaluation of future DBI frameworks.