DBus in the kernel?

[Will Thompson]

Hello!

On 05/01/10 16:52, Lennart Poettering wrote:
> IIRC the collabora folks have some data on that, i think i even heard
> something about a dbus profiler... Collaborateurs, say something!

Bustle is good for seeing extra bus traffic that you didn't expect, and
can also help you see that the time between two events is bigger than
you'd like (perhaps indicating that the daemon is being sluggish, or
maybe you're swapping like crazy, or something), but not really tell you
why.

We worked on reducing latency in message dispatch in the daemon (largely
from the POV of making your N900 ring as fast as possible). I can't find
the C profiling results right now (dead bug tracker...) but we got a
significant (as in ~20%) latency decrease by improving the per-message
match rule logic. See https://bugs.freedesktop.org/show_bug.cgi?id=23117
for some figures and patches. The patches aren't merged yet — though I
think I addressed all of Colin's suggestions — but they're shipping on
the N900 and seem to work. :)

As I recall, profiling also pointed the finger at message validation.
But we didn't have time to investigate further at the time: the match
rule stuff made enough of a difference, and seemed less invasive than
hacking at the deserializer. So I can't offer any non-anecdotal comments
on that.

-- 
Will