Season of KDE project on speed optimization

[Tom Gundersen]

Hi Aaditya,

On Fri, May 6, 2011 at 6:20 AM, aaditya chauhan
<aadityachauhan1 at gmail.com> wrote:
> Thanks for considering the idea about Speed Optimisation.
> I find the idea of decreasing the start-up time very interesting. We
> may achieve this by, reducing the number of dependencies like Markus
> mentioned. As of what, I understand we may also reduce the number of
> dependencies and may be convert a few Static libraries during start-up
> to dynamic libraries that are loaded at runtime, that might help us
> shave off a couple of milli-seconds.
>
> I read about Yuvraj Tomar's work on improving KDE start-up time. He
> improved the start-up time from 30 seconds to 19 seconds, and thats a
> huge improvement, but his work hasn't been released as of yet, so I
> assume that, some fine-tuning may be needed to his work, before
> integretion. I think we can still work on top of it, and shave off a
> few more seconds from start-up time. I am open to more discussion on
> this idea and would really be thankful if you could help me improve on
> the idea.
>
> Thank you very much Martin Sandsmark for considering mentoring me. It
> would be great experience working under/with you. :))))

Just a suggestion as a systemd and KDE fan:

Maybe it is worth having a look at how systemd has improved startup
speeds of system services, and try to do something similar to kde
startup? The idea there is to allow sockets/dbus activation to
synchronise daemon startup, rather than explicitly saying that one
daemon is started after another one is up.

Idealy something similar should be possible (and much easier) for KDE
apps/daemons as most desktop things are already able to do dbus
activation. The idea would be to start all apps/daemons/services
simultaneously as soon as the dbus session bus is running, and if one
app needs a service it will block in the call to dbus waiting for it
to start. As much as possible everything should be in the first
autostart phase.

This might be a very stupid suggestion, as I have not looked at the
code much, but judging from the bootlogd graph of my system boot some
apps are started a very long time after the dbus session bus is up and
running, so maybe there is room for improvement.

Cheers,

Tom