On 13/03/2018, Jason Dillaman wrote: > While fixing a bug in the fio RBD engine last week, I spent some time > generating and looking at perf on and off CPU flamegraphs for small IO > workloads. I never saw "nlock" taking up any time in the graphs for > librbd/librados, but for librbd::AioCompletion and > librados::AioCompletion, I did see ~5% of the time being wasted on > string constructors for the lock name. I have a branch where I > switched just librbd::AioCompletion and librados::AioCompletion to use > std::mutex since the value of lockdep seemed pretty minimal (if not > non-existent) for those two cases. We could kill off the string constructor overhead. Just have Mutex (or mutex_debug) store a std::string_view instead of a string. The caller would be RESPONSIBLE FOR MAKING SURE IT NEVER GOES AWAY (at least not within the lifetime of the mutex) but in most cases where we're just passing in a static string "foo"sv would do that. Basically a more C++ friendly/idiomatic way for storing a const char* and passing a string literal. We could also just store a const char* and pass a string literal. Whichever people think will be less error prone. -- Senior Software Engineer Red Hat Storage, Ann Arbor, MI, US IRC: Aemerson@OFTC, Actinic@Freenode 0x80F7544B90EDBFB9 E707 86BA 0C1B 62CC 152C 7C12 80F7 544B 90ED BFB9 -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
