| 30 Jul 2022 |
 @linus:schreibt.jetzt | so uh, I think this is the first time I've actually booted it successfully with a systemd initramfs | 13:16:38 |
| @linus:schreibt.jetzt | I have to say I'm quite underwhelmed by the performance | 13:16:45 |
|  @sandro:supersandro.de changed their display name from Sandro to Sandro 🐧. | 13:17:05 |
| @linus:schreibt.jetzt | Startup finished in 3.658s (kernel) + 22.254s (initrd) + 35.780s (userspace) = 1min 1.693s | 13:17:18 |
 @elvishjerricco:matrix.org | How long does the old initrd take? | 13:17:41 |
| @linus:schreibt.jetzt | about to check :) | 13:17:51 |
| @linus:schreibt.jetzt | feels like ~4s, but will wait for it to finish booting so I can ask systemd-analyze... | 13:21:10 |
| @linus:schreibt.jetzt | Startup finished in 18.164s (kernel) + 34.941s (userspace) = 53.105s ok it's not as bad as I thought | 13:22:02 |
| @elvishjerricco:matrix.org | That's still fairly noteworthy | 13:22:25 |
| @elvishjerricco:matrix.org | Note though that systemd initrd runs the NixOS activation script, whereas the old one just starts stage 2's PID1 with activation before hitting systemd. So you have to count the time spent before systemd starts as part of the "initrd" as far as performance measurement goes | 13:24:28 |
| @elvishjerricco:matrix.org | * Note though that systemd initrd runs the NixOS activation script, whereas the old one just starts stage 2's PID1 with activation before hitting systemd. So you have to count the time spent in activation in stage 2 before systemd starts as part of the "initrd" as far as performance measurement goes | 13:24:59 |
| @linus:schreibt.jetzt | ah ok | 13:25:17 |
| @linus:schreibt.jetzt | but that seems to make only about ~1s of difference | 13:25:29 |
| @linus:schreibt.jetzt | wait no | 13:25:41 |
| @linus:schreibt.jetzt | I'd expect userspace for systemd to be shorter then | 13:25:51 |
| @linus:schreibt.jetzt | maybe I need to collect more samples. | 13:26:07 |
| @elvishjerricco:matrix.org | No. Both ways count "systemd has started" as the beginning of that part | 13:26:09 |
| @elvishjerricco:matrix.org | And in that sense they're the same | 13:26:15 |
| @linus:schreibt.jetzt | oh right | 13:26:17 |
| @linus:schreibt.jetzt | [** ] A start job is running for NixOS Activation (18s / no limit) yeeeeaaaah maybe we can move some of the activation snippets into systemd services 😇 | 13:27:49 |
| @elvishjerricco:matrix.org | Activation is pretty simple but for some reason it takes forever on slower systems | 13:28:41 |
| @linus:schreibt.jetzt | ooh maybe things would be faster if my /etc were a tmpfs and not an SD card | 13:28:44 |
| @elvishjerricco:matrix.org | Yea. In my experience it takes forever to set up etc on an sd card | 13:29:05 |
| @elvishjerricco:matrix.org | I wanna say activation also does the channel copying thing for image based setups that copy a channel at boot, which also takes forever | 13:30:35 |
| @elvishjerricco:matrix.org | * (null) | 13:30:50 |
| @linus:schreibt.jetzt | oof | 13:31:32 |
| @elvishjerricco:matrix.org | Linux Hackerman: btw what kind of system is this? | 13:32:46 |
| @linus:schreibt.jetzt | pinecube | 13:32:56 |
| @linus:schreibt.jetzt | pine64.org/cube | 13:32:59 |
| @elvishjerricco:matrix.org | Oh holy crap I'm surprised you can get it to boot at all lol | 13:33:32 |
