{ lib, ... }: { # Enable in-memory compressed devices and swap space provided by the zram kernel module. # By enable this, we can store more data in memory instead of fallback to disk-based swap devices directly, # and thus improve I/O performance when we have a lot of memory. # # https://www.kernel.org/doc/Documentation/blockdev/zram.txt zramSwap = { enable = true; # one of "lzo", "lz4", "zstd" algorithm = lib.mkDefault "zstd"; # Priority of the zram swap devices. # It should be a number higher than the priority of your disk-based swap devices # (so that the system will fill the zram swap devices before falling back to disk swap). priority = lib.mkDefault 100; # Maximum total amount of memory that can be stored in the zram swap devices (as a percentage of your total memory). # Defaults to 1/2 of your total RAM. Run zramctl to check how good memory is compressed. # This doesn’t define how much memory will be used by the zram swap devices. memoryPercent = lib.mkDefault 50; }; # Optimizing swap on zram boot.kernel.sysctl = { # vm.swappiness - Controls kernel preference for swapping (range: 0-200, default: 60) # For in-memory swap devices like zram/zswap, values above 100 are recommended. "vm.swappiness" = lib.mkDefault 180; # vm.watermark_boost_factor - Controls aggressiveness of memory reclaim (default: 15000) # Setting to 0 disables watermark boost, preventing premature memory reclamation. # This allows fuller memory utilization before the kernel starts reclaiming pages. "vm.watermark_boost_factor" = lib.mkDefault 0; # vm.watermark_scale_factor - Controls kswapd wakeup frequency (range: 1-1000, default: 10) # A higher value triggers background memory reclamation earlier (at 12.5% memory pressure). # Value 125 means kswapd becomes active when free memory drops below 1/125 of total memory, # balancing memory more proactively to prevent sudden swap storms at high swappiness values. "vm.watermark_scale_factor" = lib.mkDefault 125; # vm.page-cluster - Controls swap readahead (range: 0-6, default: 3) # 0 means read only 1 page (2^0) at a time, disabling readahead. # For low-latency devices like zram, readahead hurts performance by fetching unnecessary data. "vm.page-cluster" = lib.mkDefault 0; }; }