[LAD] PCI MIDI jitter - comparison Ubuntu (bad) and Suse (might be ok)

[Arnout Engelen]

On Sun, Jul 11, 2010 at 04:53:14PM +0200, Ralf Mardorf wrote:
> today I compared a default Ubuntu Studio with and without the
> proprietary NVIDIA driver. 

OK, so the proprietary driver seems to yield better 'worst latency' values 
compared to nouveau. That's kind of odd, anything X-related would have a much
lower priority than the MIDI threads.

> Note that for Ubuntu Studio 2 tests failed because of time out errors

What exactly timed out? 

> even the tests that were passed with success are significantly less good, 
> than the tests with openSUSE, were I set up audio myself.
> 
> What might be the difference between Ubuntu and Suse?
 
> ------------------------------------------------------------------------
> Ubuntu Studio 10.04 amd64
> 2 x Terratec EWX 24/96 (2 single cards, but 1 virtual card)
> Frequency scaling ?
> ------------------------------------------------------------------------
> spinymouse at ubuntu:~$ uname -a
> Linux ubuntu 2.6.32-23-preempt #37-Ubuntu SMP PREEMPT Fri Jun 11
> 10:19:07 UTC 2010 x86_64 GNU/Linux
> spinymouse at ubuntu:~$ envy24control
> 0xcf00, irq 20, Master Clock int 44100
 
> ------------------------------------------------------------------------
> openSUSE 11.2 amd64
> 2 x Terratec EWX 24/96 (2 single cards, but 1 virtual card)
> Frequency scaling performance
> ------------------------------------------------------------------------
> spinymouse11.2 at suse11-2:~> uname -a
> Linux suse11-2 2.6.31.6-rt19 #1 SMP PREEMPT RT Wed Nov 18 16:59:26 CET
> 2009 x86_64 x86_64 x86_64 GNU/Linux

2 differences jump out: 

- the frequency scaling is set to 'performance' on your suse install and to 
  '?' on your ubuntu install. 
- The kernel options for your suse install has the options 'SMP PREEMPT RT', 
  while your ubuntu install has the options 'SMP PREEMPT'. In other words, it
  looks like your ubuntu kernel has 'normal' preemption, but not the -rt 
  patch.

The latter looks like a good candidate for explaining the difference. Could 
you test with the kernel from the linux-image-2.6.31-10-rt package instead of 
the kernel from the linux-image-2.6.32-21-preempt package which you seem to be
using now?


Kind regards,

Arnout