Towards Efficient Instrumentation for Reverse-Engineering Object Oriented Software through Static and Dynamic Analyses

Abstract

In software engineering, program analysis is usually classified according to static analysis and dynamic analysis. While static analysis provides inaccurate and imprecise results due to programming language's features, dynamic analysis produces more accurate and precise results at runtime at the expense of longer executions to collect traces. One prime mechanism to observe executions in dynamic analysis is to instrument either the code or the binary/byte code. Instrumentation overhead potentially poses a serious threat to the accuracy of the dynamic analysis, especially for time dependent software systems (e.g., real-time software), since it can cause those software systems to go out of synchronization. There are two ways to increase accuracy of a dynamic analysis: devising more efficient instrumentation and using a hybrid (static plus dynamic) analysis. A hybrid analysis is a favourable approach to cope with the overhead problem over a purely dynamic analysis. Yet, in the context of reverse engineering source code to produce method calls dynamic and hybrid instrumentations typically lead to large execution traces and consequently large execution overhead. This thesis is a step towards efficient and accurate information collection through a hybrid analysis procedure to reverse engineer source code to produce method calls, with the prime objective to reduce instrumentation overhead. To that aim, the first contribution of this thesis is to systematically analyze the contribution to instrumentation overhead of different elements of an existing and promising hybrid solution. Then, a second contribution of the thesis is to suggest an instrumentation optimization process with a range of different designs for those elements to reduce the overhead and select the best one for each element to optimize that solution. The resulting optimized hybrid technique, our third contribution, which potentially produces more accurate instrumentation compared to that hybrid solution for multi-thread software by reducing execution overhead by three quarters, has a reasonable efficiency to reverse engineer programs to produce method calls for multi-threaded software. A final contribution of this thesis is to suggest a set of recommendations for efficient instrumentation.