Re: Parallellizing the boot in Debian Squeeze - ready for wider testing

[Petter Reinholdtsen <pere@hungry.com>]

[Cesare Leonardi]
> If that helps, reading this thread i've set the previous variable in
> my notebook (Sid with Gnome environment). I can see no problem but the
> speed improvement is really small.

Great to see more test results. :)

> Here what i've measured, from the Grub start to the Gdm prompt, in
> either case starting from a completely power off machine:
> Without concurrency: 33 sec.
> With concurrency (try 1): 29 sec.
> With concurrency (try 2): 31 sec.

The brave can try to merge the symlinks in rcS.d and rc2.d to get
maximum concurrency using the declared dependencies, but such setup is
not really maintainable in the long run and I only use it on machines
I can reinstall after testing the effect.  Doing this exposes
incomplete dependencies, and is an effective way to find bugs in
packages. :)

Note that there are several serialization points in the boot today,
reducing the effect of concurrent booting, and one way to reduce that
problem is to move as many scripts as possible out of rcS.d/ and into
runlevels 1-5, to reduce the serialization effect of the switch from
rcS.d/ to rc2.d/ during the normal boot.

Only the scripts required to get the machine into single user should
be in rcS.d/, and all others should be in rc[1-5].d/.  When we get
that in place, Debian would finally have a single user mode that work
properly. :)

Happy hacking,
-- 
Petter Reinholdtsen