[LAD] discussing HZ (kernel timer frequency) and midi timing

Mulyadi Santosa a_mulyadi at telkom.net
Sat May 5 10:05:49 UTC 2007

Hi Florian....

> seq24 does _not_ use RTC for timing. It uses an approach based on sleep() (in 
> some way or the other), so setting the rtc-max-frequency does nothing for 
> seq24 performance. 

Seems correct. I tried to grab latest seq24 and compiled it. While 
testing to find out how it works, I grep-ed the source code and none 
indicate it has something to do with RTC programming. However, I will 
keep looking and studying the code.

FYI, I tested seq24 using this simple setup:
seq24--> ALSA port --> timidity (as ALSA sequencer client)
> To improve seq24 performance over the vanilla kernel you 
> need a kernel with the system timer frequency set back to something like 
> 1000Hz (vanilla has 200Hz, which is too unprecise for MIDI timing based on 
> sleeping short amounts of time)..
That's what I got too. A loop (?) take data array and using *sleep(), it 
waits until the right time to deliver the midi messages to related 
driver (in this case ALSA).

I found an interesting parameter, that is --stats. Seems like it counts 
how long it takes to work on a single MIDI message. Anybody knows more 
about this parameter?

NB: Once again, thank you Florian..



More information about the Linux-audio-dev mailing list