Post by @qwertydotplus@mastodon.social

2026-09-20 10:01 UTC

@etchedpixels@mastodon.social ...given that i can still experimentally replicate that behaviour on modern kernels... that behaviour is still present. to my knowledge, the kernel still makes a great effort to avoid OOM killing processes until it's well and truly screwed, thus in a situation where memory is nearly 100% filled with anonymous pages, you will inevitably end up thrashing file pages.

Replies (2)

@qwertydotplus@mastodon.social 2026-09-20 10:11

@etchedpixels@mastodon.social i've forced a test device into such a condition by opening multiple programs until RAM fills. the last time the display updated was 8 minutes ago (edit: finally became responsive again, after 27 minutes). no response to any interaction from mouse and/or keyboard, and i cannot ssh into the machine, as the connection times out. i'm fairly certain this behaviour is unavoidable, but i am open to being proven wrong.

Open ##5093873
@etchedpixels@mastodon.social 2026-09-20 10:54

@qwertydotplus@mastodon.social The no overcommit feature is disabled on Linux by default because it meant people had to write software properly. All the swapping/paging behaviour on Linux is fixable. There are algorithms that deal with it and they are well understood. Just nobody cares enough to bother. Firstly you don't allow more anonymous address space to exist that you have swap for. This fixes much of it but upsets all the sad "but I just mmap 10GB and figure it out later" people

Open ##5093874