[linux-audio-user] Kernel 2.6.x latency

[James Stone]

On Thu, Mar 11, 2004 at 02:52:08PM -0600, Jack O'Quin wrote:
> James Stone <stone1 at btinternet.com> writes:
> 
> > On Thu, Mar 11, 2004 at 06:21:15PM +0100, Frank Barknecht wrote:
> 
> > > Another thing to check would be your X server's nice value. This used
> > > to be "-10" in older Debians, but should not be set at all with 2.6. 
> 
> > How disappointing.. I just tried this, and it didn't help.. I was almost
> > sure this was going to work. I am still getting many more xruns than
> > with the low latency 2.4.x kernels though at least I am not getting
> > xruns on opening and closing windows!
> > 
> > However, on doing a top in 2.6.x I noticed that quite a few processes
> > are running with a nice of -10 (something called "event" among
> > others..). This does not appear to be the case in 2.4.x.. perhaps this
> > could be the source of the trouble.. not really sure how to fix it
> > though.
> 
> This makes me wonder if there is a bug in the scheduler.  IIUC, a
> realtime thread is *supposed* to have higher priority than any
> non-SCHED_FIFO thread, regardless of "nice" value.  
> 
> If we can nail down a case where this is definitely happening, someone
> should report it to Andrew Morton.
> 

I have just done some further testing: using my SBLive, running muse and
zynaddsubfx through jack (started with qjackctl), if I set the nice
value of XFree86 to 0, I do not get xruns when switching workspaces or
opening a new terminal, but with a nice value of -20 I get several xruns
in both situations.. I can also get xruns running jackd as root and
opening several aterms while it is connected to hydrogen.

although with nice 0 it is close to usable, I still get xruns on
starting new programs connecting to jackd, whereas with 2.4.25/low
latency this does not occur.

James.