Re:The Four Month Bug: JVM statistics cause garbage collecti...
by snewman on Thursday March 26, 2015 @07:25PM
(#9273833)
Profilers lie all the time. Usually it's not so much a bug, as a known limitation of the profiling approach. E.g. profilers that annotate methods with entry / exit code, inflate the runtime of small methods. Profiles that rely on CPU sampling, can be vulnerable to correlations with the sampling schedule. And so forth. The moral is to always do a reality check on whatever your profiler is telling you.
Re:The Four Month Bug: JVM statistics cause garbage collecti...
by kenbellows on Friday March 27, 2015 @07:44AM
(#9275684)
Just for the sake of it, that is Aaron Patterson (yes, aka Tenderlove), but it is not aaronsw, a moniker referring to the late Aaron Swartz.
https://www.youtube.com/watch?feature=player_detailpage&v=JM...
Turns on the profiling data where there was a discrepancy between CPU times and Wall times was only OS X because of a problem with a trap() call on OS X specifically, not any other platform. His moral of the story: even profilers have bugs.
https://www.youtube.com/watch?feature=player_detailpage&v=JM...
I think I am seeing a pattern today.