[LAU] Balance between performance and noise

Fernando Lopez-Lezcano nando at ccrma.Stanford.EDU
Fri Dec 10 01:13:31 UTC 2010


On 12/09/2010 03:54 PM, Fernando Lopez-Lezcano wrote:
> On 12/09/2010 12:10 PM, Robin Gareus wrote:
>> On 12/09/2010 06:39 PM, rosea.grammostola wrote:
>>> On 12/09/2010 06:35 PM, Robin Gareus wrote:
>>>> On 11/27/10 13:11, Arnold Krille wrote:
>>>>
>>>>> Set ther cpu-governor to performance (so it doesn't switch
>>>>> frequencies which
>>>>> produces xruns) but reduce the maximum allowed frequency.
>>>>>
>>>> what kernel are you running?
>>>>
>>>
>>>> 2.6.33.7-rt29
>>>>
>>>
>>> That one ^
>>>
>>> You tested it with yoshimi, hammond or other heavy patches? Foo-yc20 ?
>>
>> different toolset here, but I can go from naught to 16 jconvolvers,
>> csound, fluidsynth, ardour& jamin in a few seconds. causing the CPU
>> usage to jump up and change freq w/o x-runs.
>>
>> ..what does cause problems here is FSBus frequency scaling, I've
>> disabled that in the BIOS.
>
> Turns out that there are two parameters in my laptop's BIOS (Lenovo
> T510) that can disable both pci and pci express power management.
>
> I'll be darned! Guess what? The xruns are gone! Like magic!! Just on a
> preliminary test (I don't have all the hardware I would need here) but
> running with the "Ondemand" governor.
>
> This is a _golden_ tip. Thanks much!

Ah Fernando, always jumping to conclusions before all the evidence is in :-)

A more thorough test (reproducing almost all the stuff I used last time 
I performed live - the only thing missing are the usb controllers - 
reveals that the problem is different (but strangely related to power 
management).

Everything works fine in both cases - pci/pci express power management 
On and Off, and using the "On Demand" governor - until I redo all Jack 
connections using jmess. This is what triggers problems in jackd 
(1.9.6). Tons of messages and xruns reported. Switching to the 
"Performance" governor stops the error messages, but something is left 
in a broken state inside jackd. For example, opening Firefox triggers 
xruns (remember, I switched to "Performance" so now the cpu is running 
at top speed). There were no xruns doing that before using jmess. Now to 
the more confusing part. If I turn OFF the cpuspeed daemon 
(/sbin/service cpuspeed stop) the xruns go away. What?? :-)

So now I stop everything (all jack clients, jack itself). I restart the 
cpuspeed daemon. Then restart everything from scratch except for jmess 
and I get no xruns.

Will keep investigating... it would seem that jmess is doing something 
(or most probably not doing something) that is really upsetting jackd.

-- Fernando


More information about the Linux-audio-user mailing list