Starting the kdbus discussions

Tim Sander tim at krieglstein.org
Tue Jan 14 02:08:11 PST 2014


Hi
> Speed isn't the main benefit, see all of the other comments about why
> this work was done for those reasons.  But, it will be nice if it is
> faster.  
Huh, i thought speed was the main reason for moving into kernel space. 
The context switches imposed by dbus-deamon really hurt latency.
Just out of interest: could you point to somewhere what you main design 
targets were instead?

> I have heard reports of orders-of-magnitude speed increases on
> some workloads / benchmarks, but as I've never run them, and everyone's
> processor / system / workload for dbus is different, I encourage others
> to run their own tests if they really care about dbus performance
> issues.
> 
> Hint, almost no one cares about dbus performance, except people with
> really underpowered processors abusing dbus for things it was never
> originally intended to do.
Does arm11 with 533 Mhz count? But still on much higher powered systems the 
round trip times are annoying. Besides with the CPU's getting slower faster 
:-) (Single Thread is nearly flat), its the softwares turn. 

To the fact that no one cares about performance: Just think of the houres 
wasted for the people waiting for the login to complete. While not having 
benchmarked it i think that the dbus round trip times play a role in these
times.

> If anyone wants to post their numbers and the tests they ran, it would
> be great to see them.
If i get s.th up and running i will do some performance measurements.

Best regards
Tim


More information about the dbus mailing list