Touch image boot time is too long

[Pat McGowan]

All comments relative to testing on Mako.

The current images boots to full UI in 35 secs.
Enabling Mir makes this worse to the extent that the UI does not come up before the inactivity timeout shuts off the screen, leading the user to think the device did not boot.
With Mir enabled boot goes to 57 secs to full UI.
I noticed that with Mir enabled the sensorservice is consuming the cpu for a significant period of time.

The inclusion of the session startup will make this longer.

A reasonable target is 30-40 secs near term.

There is already a plan to have upstart tasks coordinate with android side startup, and remove the current sleeps used as workarounds.

[Pat McGowan (pat-mcgowan)]

Maguro bootchart at http://people.canonical.com/~ogra/ubuntu-touch/bootchart-mako.png

[Oliver Grawert (ogra)]

the above is a mako bootchart FTR. :)

http://people.canonical.com/~ogra/ubuntu-touch/bootchart-maguro.png has an old maguro chart from june (right after teh switch to flipped containers)

as you can see there, the HUD service is going wild, unity8 waits for it to be ready (20-25sec) before it fires up the indicators ... unity8 will not map anything to screen until they are up, this causes the 30-40sec black screen we see after the google logo.

the IO issues were obviously fixed since (as you can se in teh chart pat linked, it is not going wild on IO anymore) what is intresting though is that unity8 still seems to wait for HUD service for about 30sec, it seems to me as if the HUD itself was fixed but the wait in unity was not dropped at the same time.

[Oliver Grawert (ogra)]

low hanging fruit: there are a few sleeps in the uptstart jobs in /usr/share/upstart/sessions/ that we should be able to replace with putting proper upstart jobs and events in place (could buy us about 10sec i think)

[Oliver Grawert (ogra)]

another issue seems to be the semi zombified session dbus daemon, someone from teh UI teams should look at the log and find out what happens there.

[Pat McGowan (pat-mcgowan)]

Hud issue at bug 1234700

[Oliver Grawert (ogra)]

after cleaning up some of the session upstart jobs and making ureadahead work properly (it was only enabled in one kernel and its environment was broken) on all systems we are now at ~30+ seconds on maguro and at ~25-30sec on mako again for release.

note that images that are in readwrite mode are booting a lot slower (which makes it really hard to properly bootchart our actual boot process)