Post

fig (aka:[phil])

@bad-example.com

waow 50% less memory for a mimalloc -> jemalloc swap (yes that is under-3G-rss while backfilling)

22/7/26, 3:08 pm0 reposts60 likes

@botnana.bsky.social

that gap right after 10:55, was that a restart to swap the allocator in, or did rss actually take that long to settle back down after the swap

fig (aka:[phil])

@bad-example.com

yeah that was the restart (with git pull, recompile, etc in between)

@khanwinter.com

When’s jermalloc

I did something similar, I dont think moving lots of small events around threads is very efficient for the standard allocator at the rate of the firehose. Im also becoming partial to the idea of using memory arenas

fig (aka:[phil])

@bad-example.com

yep. i saw a memory *increase* from std -> mimalloc but it did get a performance jump. but it was just too tight on this machine. this is not a benchmark setup but i think the mimalloc -> jemalloc change did get another small perf jump though

dawn (｡•̀ᴗ-)✧

i think my older hydrant benchmarks were either faulty or maybe because stuff changed but looks similar on hydrant full backfill

Kuba Suder 🇵🇱🇺🇦

Hey, are these more or less up to date? I'm thinking about looking into Hydrant and if/how I could use it for some things hydrant.klbr.net/docs/w/conce... 90008.leaflet.pub/3mhp3t4kuw22e

dawn (｡•̀ᴗ-)✧

the bench numbers arent anymore from the blog post but the rest of it is yes

dawn (｡•̀ᴗ-)✧

well apparently you can get it down even more

fig (aka:[phil])

@bad-example.com

:o

dawn (｡•̀ᴗ-)✧

(also this box is uplink throttled but. doesnt rlly matter here)

dawn (｡•̀ᴗ-)✧

jemalloc-tuned being _RJEM_MALLOC_CONF=background_thread:true,dirty_decay_ms:1000,muzzy_decay_ms:0

dawn (｡•̀ᴗ-)✧

anyway i guess ill be changing back to jemalloc

@tom.sherman.is

is jemalloc the jacobian memalloc answer truthfully