Details
-
Wish
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
Description
In ARROW-6478, we switched back to jemallox 4.x to fix a large performance regression in micro-benchmarks which appeared when bumping the vendored jemalloc version to 5.2.0.
Since then, the issue was discussed with the jemalloc maintainers and we found out that customizing some options allowed regaining all (or most) of the performance on micro-benchmarks with 5.2.x.
Attachments
Issue Links
- links to