bachelorThesis
Estudo e implementação de técnicas de perfilamento por amostragem e análise de fluxo de controle em máquinas virtuais de processo
Fecha
2015-12-02Registro en:
NAPOLI, Otávio Oliveira. Estudo e implementação de técnicas de perfilamento por amostragem e análise de fluxo de controle em máquinas virtuais de processo. 2015. 41 f. Trabalho de Conclusão de Curso (Graduação) – Universidade Tecnológica Federal do Paraná, Campo Mourão, 2015.
Autor
Napoli, Otávio Oliveira
Resumen
Virtual machines play a big role in modern computing. Therefore it is crucial to provide higher performance implementations. A problem related to high performance virtual machine consists in efficiently emulating the instruction set of a guest architecture in a host machine. In order to achieve this, execution profiles must be collected and a control flow graph must be build in order to determine program execution behavior and serve as basis for decision making in dynamic optimization techniques. The main challenge related to collecting execution profile data is that this process incurs in considerable overhead to the virtual machine. The aim of this work was to study sampling-based profiling techniques to collect execution profile data, which are less costly then traditional instrumentation-based approaches, albeit having a performance-accuracy trade-off. Furthermore, a sampling-based technique was implemented in a Nintendo Entertainment System (NES) emulator. The most significant contribution of this work is a sampling-based profiling hot control flow graph construction algorithm named GFCGaussiano, along with its experimental implementation. This is a lowcost incremental algorithm that is based on simple gaussian measures that builds a control flow graph depicting the hot regions of a program during runtime. The algorithm achieves this by sampling just 1.5% of all instructions executed by the virtual machine and generates graphs with up to 83% similarity to the exact hot control flow graph gathered by instrumentation.