On 12/16/19 12:51 AM, Bruce Dubbs via lfs-dev wrote:
On 12/16/19 12:22 AM, Ken Moffat via lfs-dev wrote:
On Sun, Dec 15, 2019 at 11:54:57PM -0600, Bruce Dubbs via lfs-dev wrote:
On 12/15/19 7:06 PM, Ken Moffat via lfs-dev wrote:
On Sun, Dec 15, 2019 at 12:30:52PM -0600, Bruce Dubbs via lfs-dev
wrote:
https://www.phoronix.com/scan.php?page=article&item=10980xe-intel-linux
On the other hand on page 6, Systemd Total Boot Time
The best time was about 25 seconds. On my Haswell, my LFS System
V boot
time is 8 seconds.
My haswell often spends several seconds in boot with things like
unbound (for me, that is used on all systems, not as a server
package). But I haven't timed a boot recently. For systemd
desktops, the time will include both X (or Wayland) and the
display manager.
For SysV the boot times are in /var/log/boot.log.
Well, yes, the times for userspace. On my haswell too, 8 seconds
(give or take 1 second since it only hours, minues, seconds). But my
impression has always been that modern kernels take several seconds.
For recent boots. dmesg -e will give relative times and the local time.
dmesg -T would work also. That can be used to coordinate kernel boot
time and boot script times. It's late here so tomorrow I'll add
unbound and lightdm and post those times.
Also, I don't know how phoronix measures boot times. They might or
might not consider kernel boot time. AFAICT, the only difference
there between distros on the same HW would be minor kernel
configuration issues.
For me, the longest time at powerup or reboot is the system firmware
doing it's thing.
-- Bruce
Here are my boot times on my development system. This is with GDM
factored in:
renodr [ /sources ]$ systemd-analyze
Startup finished in 1.882s (kernel) + 23.266s (userspace) = 25.148s
graphical.target reached after 23.257s in userspace
I have a feeling that systemd-analyze is how phoronix is getting their
timings for startup.
renodr [ /sources ] $ systemd-analyze blame
14.272s dhcpcd@enp0s31f6.service
A majority of my startup time comes from obtaining an IP address over
dhcpcd. GDM starts in around 600ms for me.
I will note that my SBU time on my development system, which is a
Skylake CPU (i5-6600k) is 100 seconds. With 9.0-rc1, my SBU value was in
the neighborhood of 89/90 seconds (depending on the jhalfs run). Only
two things have changed that could've impacted it's time, and those were
binutils and the kernel (GCC and glibc versions are the same). I'm
leaning towards binutils because several of my other systems have seen
increases in the 6-7 second range, although it could potentially be due
to CPU vulnerability mitigations - I haven't honestly looked into it much.
--
http://lists.linuxfromscratch.org/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
