Deliverable 4: Proposed boot process is available for review (SoC2006)

Petter Reinholdtsen pere at hungry.com
Mon Sep 11 20:51:10 UTC 2006


[Henrique de Moraes Holschuh]
> This *requires* a new enough RTC, compiled in RTC support (not as a
> module -- it will mess with the system time in a non-predictable way
> due to coldplug), and the RTC *MUST* be in UTC.

Could be.  I have tested it on four x86 machines, and all of these
didn't need hwclock*.sh executing.  I always keep the hardware clock
on UTC, I guess that is a requirement.

I spoke with Bastian Blank on #debian-kernel, and he told me he
checked the kernel source, and for i386 the kernel would read the time
from CMOS during boot and write it during shutdown.  He said this had
been the case for at least the entire 2.6 series.

So when using the debian kernels and i386 hardware, it seem safe
enough to not use the hwclock*.sh scripts during boot and shutdown.
And this do have a huge impact on the boot speed, at least for the one
I benchmarked today.  16 seconds off a 44 second boot is 36 %. :)



More information about the initscripts-ng-devel mailing list