zramSwap: by default disable zSwap and set sysctls

[tamara-schmitz]

I am adding two improvements to the zramSwap module. The module can configure a NixOS machine such that it creates a zram block device which is then mounted as swap. This is useful in situations where it makes sense to avoid storing swap onto a physical storage device. The kernel starts to swap anonymous memory pages (like application data) when a lot of memory is either requested by applications and / or when files from physical storage are accessed frequently while application memory is never touched. By creating this setup, anonymous pages can be compressed in memory instead of forcing the kernel to try other memory freeing techniques like compaction and reclaiming. Applications will pretty much always waste memory for example by allocating a variable on the stack when launching and initializing but never touching it again until the program exits. Also most program data is highly compressible.

These changes are taken and / or inspired by the Archwiki page on Zram and the discussions linked in this Pop_OS! PR.

So here are the main new options I added and enabled by default:

• Add a kernel parameter that disables zswap: the idea of zswap is to keep memory pages that are intended to be swapped out to a physical disk and try compressing them in RAM instead. This is non-sense when your swap is also just compressed in RAM.
• Set sysctls to trigger swapping and hence compression much earlier: these are tunables and the value you set is an opinion and not absolute. Their defaults however are assuming swap on a physical storage device.
  • vm.swappiness = 150 getting file data even from a fast SSD is gonna be much slower than decompressing some pages, so let's swap more frequently
  • vm.watermark_boost_factor = 0 watermark boosting kicks off kswapd when there is memory fragmentation and someone needs a continous area. Even when using a swapfile this seems to maybe cause longer pauses as seen here. But because swapping is gonna lead to more allocations this seems counterproductive. Let reclaim do its thing instead.
  • vm.watermark_scale_factor = 125 again let's kick off kswapd earlier rather than wait for memory starved situations as swapping to zram will increase pressure more.
  • vm.page-cluster = 0 prefetching multiple pages from swap seems silly when we are in RAM anyway

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• Tested, as applicable:
  • NixOS test(s) (look inside nixos/tests)
  • and/or package tests
  • or, for functions and "core" functionality, tests in lib/tests or pkgs/test
  • made sure NixOS tests are linked to the relevant packages
• Tested compilation of all packages that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage
• Tested basic functionality of all binary files (usually in ./result/bin/)
• 24.11 Release Notes (or backporting 23.11 and 24.05 Release notes)
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
• Fits CONTRIBUTING.md.

---

Add a 👍 reaction to pull requests you find important.

[tamara-schmitz]

I updated the commit to add to the zramSwap description. I also set the optimized sysctls to be disabled by default.

[Artturin]

No need to have these

[tamara-schmitz]

Given that it is off by default, I guess. If someone does later add a zswap option in a different module, there ought to be a warning as @K900 suggested.

[Artturin]

The option search already shows the default value of options.

[tamara-schmitz]

Would you be okay with "When zramSwap is enabled, this defaults to true." ?

[K900]

This is the disableZswap option, which should be removed entirely.

[K900]

Maybe this is a good time to introduce a zswap module then? We could then direct users to the right options, or even add a warning.

[tamara-schmitz]

I think it would make sense to do this in a second PR.

[K900]

What I really don't like is that we have information in the zram docs that's not very relevant to zram, and we could have automation take care of this instead of relying on users to read things (which they never do).