| Monitoring and profiling programs provides us with opportunities for its further analysis and improvement. Typically, monitoring or profiling, entails program to be instrumented to execute additional code that collects the necessary data. However, a widely-understood problem with this approach is that program instrumentation can result in significant execution overhead. A number of techniques based on statistical sampling have been proposed to reduce this overhead. Statistical sampling based instrumentation techniques, although effective in reducing the overall overhead, often lead to poor coverage or incomplete results. The contribution of this work is a profiling technique that we call property-aware program sampling. Our sampling technique uses program slicing to reduce the scope of instrumentation and slice fragments to decompose large program slices into more manageable, logically related parts for instrumentation, thereby improving the scalability of monitoring and profiling techniques. The technical underpinnings of our work include the notion of slice fragments and an efficient technique for computing a reduced set of slice fragments. |
