Miha Rekar - What Are Flame Graphs and How to Read Them, RubyConfBY 2017 

Talk starts at 2:30

Simpel Explanation..love it

I have A and B both calling C. In the flamegraph I see A, B and C at the same horizontal level. I would expect to see C on top of A and another C on top of B.

Sorry, but I'm anti-flame-graphs. What I do instead is get a *_small_* number of random-time stack samples and *_look at them carefully._* For example, you show a speedup of 9.5x. That means you fixed something that was on the stack 90% of the time. So 10 random stack samples would have shown the problem on 9 of them. In fact, you would have seen the problem twice in as few as 2 or 3 samples, so you would have known *_precisely_* what the problem was. Once you know what it is and how to fix it, the speedup factor is what it is. You don't need to know the speedup factor ahead of time, but if you think you do, just take more samples. And I would point out - the stack samples have line-level, even instruction-level information, which flame graphs lose by looking only at functions, so with flame graphs you're still guessing what the problem actually is. And I would point out, there's never just one problem. The few-stack-samples method can be used after every fix, and the speedup factors multiply together. When you're done, the code is truly as fast as possible.