Conversation

Notices

1. Embed this notice
   Rich Felker (dalias@hachyderm.io)'s status on Sunday, 14-Dec-2025 22:45:15 JST Rich Felker

   RE: https://hachyderm.io/@kees/115717671911290728

   This. If not entirely disabled, swap size should be set to something roughly the size of "junk" processes that you expect never to get swapped back in, plus what your storage can reliably transfer in under 30 ms or so. Anything larger is just going to make bad things drag down your system rather than failing quick and reporting failure.

   • Embed this notice
     Rich Felker (dalias@hachyderm.io)'s status on Tuesday, 16-Dec-2025 12:37:24 JST Rich Felker

     in reply to

     • Demi Marie Obenour

     @alwayscurious I'm not entirely sure what you're comparing. If you're saying that "too little" swap will favor discarding read-only mappings (code) instead of writing out dirty mappings to swap, that'd generally be a good thing: you drop the write cost and only have a read cost, for way less than half the total roundtrip cost (because write is costlier than read) and zero cost at page-out time.

     But that's also at a fixed amount of exeeding physical memory. What happens is the more swap you have, the more you'll exceed physical memory before allocations start erroring out, and if you have lots of swap, by that time the system is bogged down to the point of being virtually unrecoverable because it's so slow.

   • Embed this notice
     Demi Marie Obenour (alwayscurious@infosec.exchange)'s status on Tuesday, 16-Dec-2025 12:37:25 JST Demi Marie Obenour

     in reply to

     @dalias Doesn’t that cause Linux to have to page out stuff like executable code instead?