DBIMAT: A Runtime Program Monitoring and Performance Analysis Tool Based on Dynamic Program Instrumentation Frameworks

摘要

Monitoring and performance analysis tools based on dynamic program instrumentation frameworks are widely used to help improve program's reliability and performance. However, almost all these tools suffer from substantial runtime overhead. This paper employs three methods, including multithread monitoring code instrumentation, buffer-based offline data analysis and a novel approach to construct dominant tree according to data flow forward analysis theory, to reduce the overhead. A prototype tool named DBIMAT is developed and comparative experiments show that the new tool can efficiently improve the instrumentation speed and notably reduce the overhead.