save-buffer commented on PR #13364: URL: https://github.com/apache/arrow/pull/13364#issuecomment-1159369450
Yes, that was my experience as well. When looking with Apple's TimeProfiler I saw a ton of 0.1% to 1% stack traces. The execution really is spread out such that it's hard to pinpoint an exact bottleneck. There is a handy tool in TimeProfiler that's called "invert call tree", maybe VTune has something similar. You can then search the inverted call tree for "free" and "Alloc" and such to see what percent of time is spent there. Here's what I saw: ``` 57.00 ms 7.0% 57.00 ms tiny_free_no_lock 51.00 ms 6.3% 51.00 ms tiny_malloc_from_free_list 45.00 ms 5.5% 45.00 ms tiny_malloc_should_clear 40.00 ms 4.9% 40.00 ms tiny_free_list_add_ptr 31.00 ms 3.8% 31.00 ms tiny_size 23.00 ms 2.8% 23.00 ms free_tiny 20.00 ms 2.4% 20.00 ms free 16.00 ms 1.9% 16.00 ms tiny_free_list_remove_ptr 16.00 ms 1.9% 16.00 ms _malloc_zone_malloc 14.00 ms 1.7% 14.00 ms szone_free_definite_size ``` This was before this change, but I think it mostly still applies. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
