Performance Data Collection: Hybrid Approach

As the complexity of embedded software systems grows, performance profiling becomes more and more important. Performance profiling of embedded software systems requires data collection with low overhead and high information completeness. Performance profiling consists of monitoring a software system during execution and then analyzing the obtained data. There are two ways to collect profiling data: either event tracing through code instrumentation or statistical sampling. Event tracing may be more intrusive but allows the profiler to record all events of interest. Statistical sampling may be less intrusive to software system execution, but cannot provide complete execution information. Our position is that data collection on embedded software systems should be performed using a hybrid approach that combines the completeness of event tracing with the low cost of statistical sampling. The following sections expand this position.