speechd-up

[Hynek Hanke]

On 30.7.2010 17:58, Bill Cox wrote:
> By the way, d-bus is insanely slow.  Each message/response query takes
> 200 microseconds on my quad-core i7 laptop.

Bill, this is an one or two orders of magnitude less than
what we are concerned with. We aim to produce, play and
stop sound in 40 miliseconds (1/25 of a second,
the default rate of keyboard autorepeat). The latency
of the audio output is hard to get bellow tens of ms.
So if we do not design a protocol which uses unreasonably
many DBUS messages, I can't imagine this would be
a problem. And it is very likely that it will be getting
better with time since so many things depend on it.

Did you make a comparison of that number with
unix sockets or shared memmory for a single call?
I further suspect a significant part of the 200us you
report will be the mean time your kernel takes to switch
processes, which is common for any IPC and not
related to DBUS code.

As I have said however, real bottlenecks seem to be
elsewhere.

Best regards,
Hynek Hanke