Subject: Re: [Jackit-devel] Re: [PD-announce] Re: [linux-audio-dev] ANN: k_jack v0.0.0.5 and Mammut v0.15
From: Paul Davis (paul@linuxaudiosystems.com)
Date: Fri Jan 24 2003 - 17:39:41 EET
>> I'm quite sure you get much more reliable performance this way, although
>> JACK doesn't take advantage of multip-period buffer configurations very
>> well. And this has its reasons. With a setup like shown above, you either
>> get a huge latency (64*1728/44100*1000=2500ms), or you get latency jitter
>> (latency varies between 1.45ms->2500ms). Both are something we'd _really_
>> want to avoid.
>>
>> Because of the above reasons, JACK is designed for two or three
>> period operation (-n 2 or -n 3). This way we can achieve both low
>> latency and avoid latency jitter. We could improve reliability by
>> sacrificing one or both of these goals, but for many of the use scenarios
>> we are aiming at, this is simply not acceptable.
>>
>
>Pd does not run with 2500ms standard latency, thats for sure. So if
>your theory is correct, and it probably gives latency jitter, it
>must do a pretty good job at making it unhearable.
no, there's likely not much latency jitter in pd either. after the
first 27 periods are filled (which happens in about 39ms), every chunk
of audio delivered by pd to the hardware will be output by the
hardware about 38ms later. the jitter is limited to about 1.45ms,
assuming that pd is roughly keeping up with the hardware. of course,
if its not keeping up, then sure, the latency can jitter anywhere from
1.45ms to the entire buffer, but i doubt that this is
true. theoretically its possible though. my guess is that a running
instance of pd with this h/w setup sees latency jitter between about
1.45ms and 8ms (4 periods).
>Wouldnt it be a good idea to have an option for jack to hide latency
>jitter? Or is it to much work (hacking) with the design model?
JACK is specifically designed not to allow latency jitter. you can't
properly sync audio+video with latency jitter, because the software
cannot know for sure when the audio it generates will actually become
audible. pd, by contrast, allows significant latency jitter but in all
likelihood doesn't cause it to wobble all that much.
However, as Kai pointed out, nobody runs JACK with -p 64 -n 27, and if
they did, I suspect that they'd get similar performance to pd. it
would be interesting to hear how pd works with, say, a period size of
1024 and a buffer size of 2048 (i.e. 2 mid-size periods) instead of
lots of small periods.
--p
This archive was generated by hypermail 2b28 : Fri Jan 24 2003 - 17:35:19 EET