On Mon, Jun 14, 2010 at 09:29:58AM +0200, Troels Henriksen wrote: > Anselm R Garbe <garb...@gmail.com> writes: > > > Regarding the boot speed I disagree. I think short boot cycles can be > > achieved with rather more simple init systems than the insanity people > > got used to like the SysV style Debian insanity. A simple BSD init > > based or even more simple system always outperforms any "smart" > > technique in my observation. > > Well, for really excellent performance, you do need the ability to > parallelise the init operations, so that's a bit of complexity that has > actual performance benefits. > > I agree there is little value in the general runlevel mess.
I fully agree. after looking to minit & stuff, I decided to write our own init daemon to incorporate some safety stuff. * booting is done in parallel. * udev (+/- 5sec) was replaced by our (small) fdev (now takes some 0.1 sec). some examples: dell laptop: booting was over 45 seconds (from kernel starting timers), now 15. via epia board: was 25, now 4.3 seconds embedded ARM cpu: (never used debian there, but busybox): no final measurements, but boottime of 18 seconds got reduced to 6. OpenMoko: boottime is originally (very long) 2m40s, reduced to 35. I admit our init is quit more complex than strictly necessary (we try to guarantee that a watched process is not dead-locked, and therefore have a hardware watchdog in the init process, and ...). I'm not familiar with BSD init's. Kurt > > -- > \ Troels > /\ Henriksen >
