Boot sequence speed-up?

Ken Moffat ken at
Sat Feb 3 11:17:00 PST 2007

On Sat, Feb 03, 2007 at 10:45:58AM -0800, Dan Nicholson wrote:
> On 2/3/07, Randy McMurchy <lfs-user at> wrote:
> > Dan Nicholson wrote these words on 02/03/07 12:11 CST:
> > >
> > > This is actually something I want to bring up. Our booting is dog
> > > slow. Maybe it's time to look into making improvements. We could
> > > replace init with init-ng or upstart. Or, we could just work to
> > > parallelize the bootscripts like is done on RedHat and SuSE. I think
> > > this has been brought up before.
> >
> > "dog slow". I suppose you'd have to enlighten me how much faster this
> > old 500mhz system would boot if the boot sequence was "improved". On
> > my last boot, the logs show this:
> Mostly what I mean is that from the time boot has started to getting a
> prompt. If you have a laptop, for example, you're booting frequently
> and 25 seconds is kind of long. Or, at least, 25 seconds is much
> longer than it needs to be. Point is taken, though, and maybe "dog
> slow" isn't the right phrase.
> At the minimum, things could be vastly sped up by not serializing the
> whole operation. Read this article from IBM for an example.
 I've seen these things, they all sound absolutely wonderful. I
do boot my desktops frequently, and into gdm - nowadays they are
almost always switched off when I'm sleeping, and I might use two or
three different machines during the day. My ibook OTOH is only
booted infrequently (it sleeps beautifully).  BUT, I've looked at
this before from a high level - wait for the bios or OF to run
whatever checks it can and then present the bootloader; select the
desired kernel/system ; wait while udev, sysfs, and perhaps some
machine-dependant slow low-level things bring the kernel up, then
watch userspace.  For me, the big delays in bringing up userspace
are ntp, and to lesser extents dhcp, starting cups (varies -
sometimes seems very slow, probably version dependant), and mounting
an nfs share.  Oh, and starting X itself - it doesn't matter how
fast the machine gets, X always takes too long to start (several
seconds, feels like an eternity!).

 Sure, for my desktops I could save time by running ntp in the
background, but I like the comfort (or not) of seeing how far out
the time was.  The mounting, in my case, could be in the background
but I definitely want to know if it fails.  My best guess is that
from power-on to graphical login takes towards a minute.  If I save
10 seconds, it will be nice, but it's not that much of a difference.
Still, I await your initial hint with interest ;)

das eine Mal als Tragödie, das andere Mal als Farce

More information about the lfs-dev mailing list