Investigation Upstart/Sysvinit

[Alessandro Iurlano]

On 10/26/07, Marcin Juszkiewicz <openembedded at haerwu.biz> wrote:
> Dnia piątek, 26 października 2007, Graeme Gregory napisał:
>
> > Also when upstart gains parellel starting capability I think this will
> > allow use to get to X much earlier in the boot process. Showing people
> > the prettiness of the GUI much earlier than in sysvinit. Services can
> > then still be starting in the background. Im not sure this is what we
> > want but its a possibity.
>
> Even now you can start X11 sooner (not on first boot anyway). We have many
> services running which are not needed to start X11 (portmap, networking
> etc).
>

I have done some analysis on the boot process of GTA01. They are shown
in this page:
http://wiki.openmoko.org/wiki/Bootcharts

I have also created a lighter version of bootchart package that
substitute most of the bash script statistic collection (very spawn
process intensive) with a very light C program, significantly lowering
CPU usage of the bootchart tool. The package is on my repo and its
name is litebootchart.
You could use this to verify how upstart performs on the Neo.

I think however that, at least in GTA01, the bottleneck is the CPU as
it showed always to be at 100%.
It could also be that context switching on this processor is
expensive. If this is the case, booting in parallel could worse the
overall load time significantly.

The bootcharts don't show I/O stats as the mtd driver does not publish
that information in /proc/diskstats file. I have tried to add it
myself but could not find any source of info on how that stats are
added to diskstats (I have never done linux kernel hacking but I'd
like to start). Does anyone know where I could find this kind of
information?

Hope this helps,
Alessandro