Skip to content

LordOS — a free, fast and secure GNU/Linux distribution based on Debian, with its own desktop environment: display server and Wayland compositor in C, a React UI engine on QuickJS, and private native apps.

5.6 KB
Raw Blame

Versioned results

Measurements and checks are published as files, per LordOS version and per machine, with the commands that produced them, so anyone can repeat them and compare. A screenshot proves one moment; these files can be checked.

results/<version>/<machine>/
    README.md        the conditions (hardware or VM, monitor or headless, idle time, apps open, power)
                     and the exact commands
    sysreport.json   lordos-sysreport report --json: the hardware and what LordOS made of it, with no
                     serial numbers, addresses or user and host names (tests/sysreport_test.c checks it)
    testbench.json   lordos-testbench --gl --auto N: the Wayland checklist, frame times (average, p50,
                     p99, frames that missed the refresh budget), presentation, the 3D renderer, latency
    memory.txt       the whole system (`free -m`) and PSS per process (/proc/<pid>/smaps_rollup)
    boot.txt         systemd-analyze, and the time from the kernel's start to the desktop's first frame
                     (lord-sessiond's log line; the boot splash shows its own timer from a little later)
    lordbench.json   from 0.8.6, LordBench (docs/BENCHMARK.md)

tools/results/collect.sh <machine> [--no-memory] [--bench-seconds N] [--version V] writes such a folder; run it as the user inside a LordOS session, and with sudo to include root's processes in the PSS list. Fill in the conditions in its README before committing. Each release attaches its results/<version> folder to the GitHub release (docs/RELEASING.md).

Boot time: check it yourself

LordOS measures every boot the same way, on any machine, with no tool to install:

  • On the screen. The boot splash shows a timer in its bottom right corner, in seconds and milliseconds, from the moment the splash starts. About LordOS → Startup time shows this boot's result: the seconds from the kernel's start to the first frame on the monitor, and whether that frame was the login screen or, with automatic login, the desktop.

  • In the log. lord-sessiond writes one line per boot when the first frame reaches the monitor:

    journalctl -b -u lordos | grep "first frame"
    # first frame on the monitor 7.224 s after the kernel started (LordOS 0.8.5~rc1-beta, build 20261026, session)
    

    -b -1, -b -2… give the previous boots while the journal keeps them.

  • The history. Every boot is also appended to /var/lib/lordos/boot-times.jsonl (root's, readable by everyone, the last 200 boots), one JSON object a line, so boots and versions can be compared after the journal is gone:

    cat /var/lib/lordos/boot-times.jsonl
    # {"bootId":"e519ea86-…","at":1793052001,"seconds":7.224,"shown":"session","version":"0.8.5~rc1-beta","build":"20261026","kernel":"6.12.113+deb13-amd64"}
    python3 -c 'import json; [print(f"{d[\"version\"]:>16} {d[\"seconds\"]:7.3f} s  {d[\"shown\"]}") for d in map(json.loads, open("/var/lib/lordos/boot-times.jsonl"))]'
    

    seconds counts from the kernel's start (CLOCK_BOOTTIME), so the firmware and the boot loader are not included; shown is login (the login screen), session (automatic login) or none (no monitor).

  • The pieces. systemd-analyze splits the boot into kernel and userspace, and systemd-analyze critical-chain lordos.service shows what LordOS waited for. The splash's own timer starts when Plymouth starts, a little after the kernel: journalctl -b -o short-monotonic | grep -E "kernel: Linux version|Started plymouth-start" gives both moments, and the difference.

  • Comparing fairly. The same machine, the same settings (automatic login on or off changes what the first frame is), several boots of each version (the first boot after an update or a new kernel is slower: it builds caches); report the median. tools/results/collect.sh saves all of the above in boot.txt and boot-times.jsonl.

Memory: two magnitudes

free's "used" is the whole system's memory (kernel, Debian's services, LordOS's services, the session), without reclaimable page cache: this is the number to compare between systems, on the same machine, after the same idle time, with the same apps open. A process's RSS (ps, /proc/<pid>/status) counts every shared library page it maps, so RSS figures of several processes overlap and can't be added; PSS divides each shared page among the processes that map it, so PSS figures add up. The paper's §7.1 keeps the two apart.

Results so far

Version Machine Conditions What
0.8.5-beta vmware-fusion the same VM, updated from 0.8.4.1 to the 0.8.5 package with LordOS's package engine (lordos.org/apt-staging), measured about five minutes after the restart, Chrome, VS Code, Spotify and llama.cpp installed but not running whole system 592 MB used (free -m); 59.8 fps at 60 Hz, p50 16.6 ms, p99 17.9 ms, 1 frame over budget in 30 s; 16 Wayland checks passed; first frame 7.4 s after the kernel started (systemd-analyze 7.6 s)
0.8.5~rc1 vmware-fusion VMware Fusion VM, virtual GPU (vmwgfx, SVGA3D), 1280×800 @ 60 Hz, updated from 0.8.4.1 by apt install whole system 606–622 MB used one minute after sign-in, PSS of all processes 316 MB (login screen 78 MB of it); 59.8 fps at 60 Hz, p99 18.4 ms; 17 Wayland checks passed; boot 7.5 s
0.8.5~rc1 vmware-fusion-before-pacing the same, before software vblank pacing 524 fps at 60 Hz: the driver completes flips at once
0.8.5-dev vm116-headless Proxmox VM, no GPU, headless session (remote desktop only), development install sysreport, testbench (16 Wayland checks passed; llvmpipe; no frame presented without a monitor), boot