On Friday, June 23, 2017, Adam Williamson
<adamwill(a)fedoraproject.org>
wrote:
> So, funny story: my todo list has this item, which I'm crossing off
> with this mail:
>
> compare f21 / f22 / f23 boot on rust:
>
https://blogs.gnome.org/mcatanzaro/2016/01/31/on-boot-times/#comment-120
>
> It's only been on my list since 2016-02-06, no big deal, right?
> However, never let it be said that I'm a liar! I *did* finally get
> around to doing it, plus I included F24, F25 and the current F26
> nightly compose as well.
>
> I tested each by installing from the x86_64 Workstation live image to
> my test box, which has a low-end AMD CPU (A4-7300), 8GB of RAM, and a
> Seagate 500GB SATA hard disk (not an SSD). I created a user during the
> install (so first boot wouldn't hit gnome-initial-setup before gdm),
> then measured the boot time from hitting 'enter' on the bootloader
> screen to GDM appearing fully loaded. I then measured the time it took
> from hitting 'enter' on the password prompt to GNOME appearing fully
> loaded. Then I rebooted and did the same measurements again (to see if
> second boot differed significantly from first boot). I did not use
> encryption.
>
> The results were in contrast to mcatanzaro's from that blog post. I
> found no significant difference in boot times across F21-F25. With only
> one exception, pretty much every 'boot to GDM' measurement came out at
> 30 seconds, give or take a second or two, and every 'GDM to GNOME'
> measurement came out in the range of 6 to 8 seconds.
>
> The exception was the first boot of Fedora 21, which took a little over
> a minute. The second boot took 29 seconds, though, in line with all
> other tests.
>
> However, F26 does seem slightly slower than F21-F25. First boot to GDM
> for today's F26 nightly measured at 42 seconds. Second and third boot
> measured 39 seconds. GNOME load time was 8-9 seconds. Looking at the
> systemd-analyze info, I don't see any obvious single 'culprit', it
> seems more like approximately the same stuff is happening during boot,
> it's just all taking a bit longer.
>
> So I'm *still* not sure what's behind the very slow F23 boot Michael
> reported, but my test couldn't reproduce that slowness or show any
> significant difference between any of the releases from F21 to F25. The
> difference between F25 and F26 might be worth some independent re-
> testing and deeper analysis.
>
Test with a non debug kernel?
The 26 kernel is not a debug kernel, and hasn't been since 4.11.0-1 on
2017-05-01. Debugging is usually disabled permanently somewhere between
Alpha and Beta in each cycle.
--
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net