Meta’s Renewed Commitment to jemalloc
33 points by gaffneyc
33 points by gaffneyc
TLDR; Meta is once again investing into making jemalloc a general purpose allocater and not just built for their purposes.
Related: https://jasone.github.io/2025/06/12/jemalloc-postmortem/
What does this coded language mean?
In recent years, there has been a gradual shift away from the core engineering principles that have long guided jemalloc’s development.
probably just that they started rushing code out the door instead of taking the time to get it right, and that the accumulated technical debt finally caught up with them.
IIUC, Jason Evans got angry at the way Meta tried to catch up (https://jasone.github.io/2025/06/12/jemalloc-postmortem/#:~:text=of%20many%20others.-,Phase%203:%20Meta,-The%20nature%20of) to tcmalloc-style huge page support (https://www.usenix.org/system/files/osdi21-hunter.pdf). My guess is someone had "Land HPAA" as an OKR and did everything to hit it, or something like that (landing temeraire was also a bit of a shit show. fleet-wide allocators are hard)
The sad part is jemalloc's old chunk allocator interface is pretty amazing for 1G pages, so, to me, it feels like trying to catch up to last decade's hot thing.
Does anyone know about a proper discussion they can link to for ARM related performance point in the article?
Quick search issues on github seems to be resolved by now