[LAU] Realtime latency kernel testing

[Robin Gareus]

On 01/07/2011 10:37 PM, Gabriel M. Beddingfield wrote:
> 
> 
> On Fri, 7 Jan 2011, torbenh wrote:
> 
>> On Fri, Jan 07, 2011 at 08:05:59AM -0600, Gabriel M. Beddingfield wrote:
>>>
>>> Indamixx 2/Transmission 5 (MeeGo) has a PREEMPT kernel, and
>>> therefore rtirq won't work with it.  We've spent a lot of time
>>> tuning this kernel, and while there have been no clinical tests...
>>> we're experiencing performance very close to Transmission 4's
>>> PREEMPT_RT kernel (With jack running at 512x2 frames/buffer and
>>> 44100 Hz) on various machines.
>>
>> ahh... 512 is pretty high latency, normal PREEMPT kernels should do that
>> pretty well.
> 
> I knew _someone_ would chime in about 23ms being "high latency."  :-)

LOL, but it is.

As far as I'm concerned personally, when using the computer as
effect-rack playing guitar or bass:
latencies between 10 and 20ms are the worst!

 >20ms I can manage and compensate somewhat.
 <10ms is fine.
 <5ms is great.

I do run jack at 64*2 - jack_delay measures 4.9ms total round-trip
latency. It rocks!

> FWIW, several popular apps (esp. those doing Fourier Transforms, like
> time-stretchers) will not work reliably at <256 frames/period.

Would you care to enlighten us which popular apps that are?
'fmit' for example uses FTT and works just fine at 32fpp.

2c,
robin