- 79comments
- 165comments
- 3comments
- 26comments
- 213comments
- 54comments
- 10comments
- 40comments
- 235comments
- 70comments
- 234comments
- 3comments
- 13comments
- 49comments
- 28comments
- —discuss
- 14comments
- 160comments
- 19comments
- 172comments
- 134comments
- 88comments
- 59comments
- 12comments
- 261comments
- —discuss
- 119comments
- 25comments
- 15comments
- 16comments
Currently, it's unclear why - I didn't find any information in their public channels.
They un-archived it today long enough to revert a change to the purging algorithm that was merged in March [0]. The first archive date I saw this morning was from the last week of May. I won't speculate about the reason, but it's a small relief that the archiving seems deliberate and that the maintainers are still active.
[0] https://github.com/jemalloc/jemalloc/commit/27d7960cf9b48a9a...
More: https://github.com/facebook/jemalloc/discussions/7
We once went through the effort to use jemalloc for workloads which were highly optimized particularly for NUMA.
Incredibly, it had absolutely zero performance improvement. I’ve always wondered how that was possible. Spent a fair bit of time confirming the memory layout and NUMA node ownership etc. It was working, it just.. did nothing!
Why? The prior conversation (linked here in this posting thread) has somebody asking the same question. And a comment: did nothing. So it could be the variant has low enough achievement over other *alloc that it was felt best to retire it.
Performance of jemalloc is good, hope to know why it is archived
Looks like it got forked to https://github.com/facebook/jemalloc