Boot timing (sysv services)

[vcaron]

I've tried a very simple hack to time the different service startup. I love Debian and its /etc/default/* shell fragments, they allow lots of useful hijacking:

# Append to your /etc/default/rcS

debug_timing() {
    case "$1" in
        splash*) $@;;
              *) { time $@; echo $@ >&2; } 3>&2 2>&1 1>&3- | (tr '\n' ' '; echo) >>/lib/init/rw/timing.log;;
    esac
}
export debug=debug_timing

Forgive the shell ugliness only to format one service per line. Result on my Neo:

 real   0m9.203s user   0m0.420s sys    0m1.165s /etc/rcS.d/S03udev start 
 real   0m0.358s user   0m0.170s sys    0m0.145s sh /etc/rcS.d/S04mountdevsubfs.sh start 
 real   0m0.225s user   0m0.075s sys    0m0.110s /etc/rcS.d/S05bootlogd start 
 real   0m1.210s user   0m0.260s sys    0m0.360s sh /etc/rcS.d/S10checkroot.sh start 
 real   0m1.639s user   0m0.090s sys    0m0.200s sh /etc/rcS.d/S11hwclock.sh start 
 real   0m1.245s user   0m0.425s sys    0m0.705s sh /etc/rcS.d/S12mtab.sh start 
 real   0m0.306s user   0m0.145s sys    0m0.140s /etc/rcS.d/S18ifupdown-clean start 
 real   0m2.528s user   0m0.190s sys    0m1.115s /etc/rcS.d/S20module-init-tools start 
 real   0m1.469s user   0m0.230s sys    0m0.700s sh /etc/rcS.d/S30checkfs.sh start 
 real   0m0.265s user   0m0.100s sys    0m0.150s /etc/rcS.d/S30procps start 
 real   0m0.434s user   0m0.125s sys    0m0.240s sh /etc/rcS.d/S35mountall.sh start 
 real   0m0.689s user   0m0.200s sys    0m0.455s sh /etc/rcS.d/S36mountall-bootclean.sh start 
 real   0m0.109s user   0m0.025s sys    0m0.060s /etc/rcS.d/S36udev-mtab start 
 real   0m0.395s user   0m0.115s sys    0m0.180s /etc/rcS.d/S39ifupdown start 
 real   0m1.415s user   0m0.595s sys    0m0.720s /etc/rcS.d/S40networking start 
 real   0m0.818s user   0m0.220s sys    0m0.545s /etc/rcS.d/S50alsa-utils start 
 real   0m0.987s user   0m0.325s sys    0m0.635s sh /etc/rcS.d/S55bootmisc.sh start 
 real   0m0.317s user   0m0.140s sys    0m0.150s /etc/rcS.d/S55urandom start 
 real   0m0.371s user   0m0.185s sys    0m0.165s /etc/rcS.d/S70x11-common start 
 real   0m0.091s user   0m0.020s sys    0m0.050s /etc/rcS.d/S99stop-bootlogd-single start 
 real   0m0.875s user   0m0.200s sys    0m0.640s /etc/rc2.d/S12dbus start 
 real   0m0.873s user   0m0.045s sys    0m0.245s /etc/rc2.d/S20dropbear start 
 real   0m0.939s user   0m0.160s sys    0m0.660s /etc/rc2.d/S20gpsd start 
 real   0m0.221s user   0m0.085s sys    0m0.105s /etc/rc2.d/S20nodm start 
 real   0m1.171s user   0m0.120s sys    0m0.470s /etc/rc2.d/S22gsm0710muxd start 
 real   0m0.414s user   0m0.090s sys    0m0.055s /etc/rc2.d/S99rc.local start 
 real   0m0.295s user   0m0.035s sys    0m0.070s /etc/rc2.d/S99rmnologin start 
 real   0m0.373s user   0m0.105s sys    0m0.090s /etc/rc2.d/S99stop-bootlogd start 

Winner hands down is udev. Since OpenMoko? is fixed and known hardware, we could provide a static /dev ? Then udev could be started much later to handle pluggable thingies.

[jblondon]

Hi,
thank your for this bench, khorben what is your opinion about that ?

[Detructor]

isn't that already deployed in hackable:1? It seems so...

[Deubeuliou]

to be confirmed
zecrazytux, are you interested by this one ?

[zecrazytux]

Bootchart there:  http://fuckpaste.zecrazytux.net/h1/bootchart.png

[zecrazytux]

 http://blog.hackable1.org/