Re: look at this: samsung nc10 bootchart with mobil V2

[Simo Sorce <ssorce redhat com>]

On Thu, 2009-01-29 at 22:54 -0500, Bill Nottingham wrote:
> Brian Maly (bmaly redhat com) said: 
> >> My question is simply: How do they (the intel developers) reach that
> >> boot time without removing all those comfort services (udev etc.)?
> >>   
> > (1) its UEFI
> 
> 1) I've *NEVER* seen an EFI BIOS that wasn't abysmally slow
> 2) bootchart doesn't measure BIOS speed anyway
> 
> > (2) rc stuff is parallelized (/rc_parallel/="YES") during the startup of  
> > the rc scripts
> > (3) /boot lives on the solid state disk and root fs runs in ramdisk
> 
> It's not tricks - if you really want, there should be data from Arjan
> and Auke's plumber's conference talk.

In the fedora chart there seem to be a huge time (form sec 2.5, to sec
15) where the only things running are polymouthd and nash-hotplug, then
another gap when udev starts, then everything start booting more or less
in parallel.

In moblin there is no plymouthd/nash-hotplug and udev seem not to
introduce delays.
What are plymouthd/nash-hotplug doing ? (if its their fault)
Why fedora's udev seem to slow down another bit while moblin's udev seem
not ?

Simo.

-- 
Simo Sorce * Red Hat, Inc * New York