The On-The-Fly-Critical-Path-Tool (OTF-CPT) can already perform the critical path analysis online during the execution of the measured application. However, the analysis is limited to only track the execution time of the critical path, which often is not enough context to understand the observed performance of an application. This thesis extends the OTF-CPT by introducing a new interface to allow fortracking arbitrary metrics on the critical path. The new interface is then implemented to support the tracking of hardware counters using PAPI. An evaluation is done that investigates the overhead of the added capabilities and compares the results from OTF-CPT with results from other tools, like Score-P and Scalasca. The evaluation shows that although the overhead can vary depending on the characteristics of the measured application and how many metrics are being tracked, it is acceptable in most cases. It also shows that the results obtained using the new metrics are comparable to the results from other tools and how they could be used to characterise the performance of an application.
Leon Streichardt (Thu,) studied this question.