U-Boot slow "uncompressing"

Carsten Haitzler (The Rasterman) raster at openmoko.org
Tue Jan 22 00:43:46 CET 2008


On Tue, 22 Jan 2008 02:25:54 +0530 "pgssgp at gmail.com" <pgssgp at gmail.com>
babbled:

> Hi,
> >> Somebody in the thread at some point said:
> >>     
> >
> >   
> >>>> In OE we're no longer using the u-boot uncompressor, but use the
> >>>> one built in in the kernel, which is way faster.
> >>>>         
> >>> Wah thanks Mickey, I'll change over... that explains it.
> >>>       
> >
> >   
> >> Yep chopped decompression down to ~1s... great...
> >>     
> >
> > Good.
> >
> >   
> >> It's ~18s from power -> rootfs mounted and init running now... that
> >> includes putting the splash up but with loglevel=0 so there are no boot
> >> messages on the LCM.
> >>     
> >
> >   
> >> If we set init=/bin/sh on the kernel commandline, we can actually meet
> >> this 20s boot time target... kinda....
> >>     
> >
> > Well, 20s is almost insanely tough. My N810 takes 40s.
> >
> >   
> Just to add, as jffs2 is used as  root filesytem  which by default does 
> compression .
> And while executing an application it does decompression this will 
> increase application launch time.
> To avoid this we should disable jffs2 compression (i think there is an 
> option to mkfs.jffs2), don't know
> how this will be reduce the boot time

given our really slow flash speed of 2-3MB/sec - i suspect compresion speeds
things up as the IO is the bottleneck, not the decompression... but has someone
got benchmark numbers to back this up? i'm just suspecting. even so - i dont
think it will be a huge gain - if at all as we win on cpu and lose on having
our output IO capped at 2-3MB/sec.


-- 
Carsten Haitzler (The Rasterman) <raster at openmoko.org>




More information about the openmoko-kernel mailing list