Recent searches

No recent searches

Search options

Only available when logged in.
wetdry.world is one of the many independent Mastodon servers you can use to participate in the fediverse.
We are a community focused on gaming, tech, entertainment, and more.

Administered by:

Server stats:

729
active users

wetdry.world: AboutStatusProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.5.0-alpha.2+chuckya

...i think i may have finally figured out a root cause of the esmbot image server memory issues

and it's not even my fault

··Web
0boosts·15favorites·0reactions

this has been an unsolved issue for years

i have managed to work around it by lowering the amount of arenas and occasionally calling malloc_trim(0)

@esm@wetdry.world this one's lukewarm take of glibc being more or less abandonware with occasional patches is further proven once again

@esm excited for the new series from Essem Productions, the Broadway musical "I Hate The Glibc C Library"

@doskel @esm

It shall be complemented by "I love musl libc"

@alexia @doskel @esm keep in mind the major catch with musl is it's much slower

@niko @doskel @alexia @esm is it that much slower though? i thought the only major problem was the allocator and if you're doing something where the allocator matters you'd already be preloading jemalloc or mimalloc

@kopper @niko @alexia @esm yeah, i use chimera which is musl and afaik they have a few small patches to the allocator to bring it nearly inline with glibc anyway

@doskel @alexia @esm @niko chimera switched to mimalloc for their system alloc i believe

@kopper @niko @doskel @alexia it's funny because in this specific scenario jemalloc/mimalloc actually ended up being worse after applying the mitigations

@kopper @doskel @alexia apparently TLS is much slower with musl

@niko @doskel @alexia it might very well be. looking up musl performance stuff is extremely hard because the search results are littered with docker people going "you have to recompile pip packages for musl so it's slow and bad!!!1"

@kopper @doskel @alexia string functions in glibc are handwritten asm for arm64/x86_64 and variants and in musl they're just plain C (much slower)

*

@kopper @doskel @alexia apparently memcpy and friends are also an order of magnitude slower on average for the same reason :neofox_0_0:​

@alexia @doskel i'm not exactly a huge fan of musl either

@esm @doskel what about llvm-libc? bionic maybe if you're evil?

@alexia @doskel i actually tried experimenting with llvm-libc the other day only to find it can't be built in standalone/full mode as a shared library

i have not done much with bionic

TrendingLive feeds
About