[asterisk-users] Terrible dahdi_test results

[Mike Leddy]

I remembered I have an older box with a Wildcard TE12xP that 
uses the wcte12xp module with a newer 3.9.11 kernel that works 
perfectly.

I setup the problematic machine with the same kernel in the hope 
that this might be relevant. Unfortunately the same situation 
persists.

I used the /proc/timer_stats to see how the timers were used:

Timer Stats Version: v0.2
Sample period: 10.002 s
....
  311, 15081 kworker/u:0      mod_timer (te12xp_timer)
....

With the TE110P I couldn't find any entry.... It seems that the
timing mechanism is different, it doesn't use mod_timer.

I'm running out of ideas. Please help.

Thanks,

Mike

On Tue, 2014-05-13 at 17:56 -0300, Mike Leddy wrote:
> Thanks again Russ,
> 
> Just a quick reply for now.
> 
> No virtualization, but yes I am running a tickless kernel:
> 
> #
> # Processor type and features
> #
> CONFIG_NO_HZ=y
> 
> Standard for debian kernels. I booted with nohz=off and the behaviour
> changed. Unfortunately for the worse:
> 
> # dahdi_test 
> Opened pseudo dahdi interface, measuring accuracy...
> 66.653% 66.683% 66.683% 66.807% 67.705% 66.666% 66.651% 66.679% 
> 67.516% 66.882% 66.649% 66.657% 66.678% 66.668% 66.672% 66.664% 
> 66.675% 66.675% 66.659% 66.692% 66.631% 66.187% 66.650% 66.710% 
> 66.648% 66.633% 66.714% 66.638% 66.688% 66.794% 66.645% 66.696%
> --- Results after 32 passes ---
> Best: 67.705% -- Worst: 66.187% -- Average: 66.726523%
> 
> Comparing the boot messages without nohz=off:
> 
> [    0.000000] hpet clockevent registered
> [    0.000000] Fast TSC calibration failed
> [    0.000000] TSC: Unable to calibrate against PIT
> [    0.000000] TSC: using HPET reference calibration
> [    0.000000] Detected 2593.456 MHz processor.
> 
> and with nohz=off:
> 
> [    0.000000] hpet clockevent registered
> [    0.000000] Fast TSC calibration using PIT
> [    0.000000] Detected 2593.225 MHz processor.
> 
> I am encouraged that we seem to be homing in on the problem. I need to
> read up a bit more on the subject.... and look at possible power 
> saving issues on this machine.
> 
> Best regards,
> 
> Mike
> 
> 
> On Tue, 2014-05-13 at 15:26 -0500, Russ Meyerriecks wrote:
> > On Tue, May 13, 2014 at 7:28 AM, Mike Leddy <mike at loop.com.br> wrote:
> >         But on examination the /etc/init.d/dahdi start was only
> >         loading
> >         the dahdi module.
> > 
> > 
> > With this in mind I might start looking around the system for things
> > which might cause jitter in the servicing of system timer interrupts:
> > 
> > 
> > Are you running under a virtualized environment?
> > Are you running a tickless kernel? (maybe try adding nohz=off to your
> > kernel boot parameters)
> > Is there some sort of processor power saving or frequency scaling
> > going on that interrupts the system timer?
> > 
> > 
> > -- 
> > Russ Meyerriecks
> > Digium, Inc. | Linux Kernel Developer
> > 445 Jan Davis Drive NW - Huntsville, AL 35806 - USA
> > direct: +1 256-428-6025
> > Check us out at: www.digium.com & www.asterisk.org
> 
> 
>