Everybody hates a lying hobbit…  But are profilers broken or are we naive in our expectations?  In this talk we will re-visit some broken assumptions and try and get to grips with the limitations of our tools. We’ll cover:

— Fiction of instruction level profiling  — Skid and PEBS

— Safepoint bias, theory and practice

— JMC blindspots, theory and practice

— Honest-Profiler vs. Async-Profiler

— Profilers vs. Inlining and Inlining vs. Programmer 

— Profilers vs. Multi-threading