[LAU] jdelay - measured latency of a usb interface

[hollunder at gmx.at]

On Tue, 24 Mar 2009 11:14:00 +0100
<hollunder at gmx.at> wrote:

> Hi everyone.
> I was kind of curious about the real latency of my system and did a
> series of tests using jdelay (http://www.kokkinizita.net/linuxaudio/).
> 
> I share my results so others can see what to expect from such a usb
> audio device and also because..
> 
> ! Important !
> I ran in a number of oddities that I can't explain myself. This could
> mean that my results here are totally useless, so please don't take
> them for real until someone can hopefully clarify those matters.
> 
> 
> General Information:
> -------------------
> 
> Audio Interface: Edirol UA-25
> http://www.rolandus.com/products/productdetails.aspx?ObjectId=704
> http://alsa.opensrc.org/index.php/Edirol_UA-25
> 
> It can work in 44.1, 48kHz duplex and 96kHz either record or playback,
> so I couldn't (and wouldn't have anyway) tested 44.1 and 48kHz.
> 
> There are Periods/Buffer from 2 to 42 possible, most common is I
> believe 2 and 3 Periods/Buffer, I only tested these.
> 
> My pc is rather old, mainboard is a via km400 or km400a if that helps
> anything.
> 
> On the software side there is Arch Linux with it's latest stock
> kernel, 2.6.28.8, jack 0.116.2 and qjackctl 0.3.4 (the patchbay made
> this a lot easier).
> 
> Measurement Results:
> -------------------
> 
> SR 44100
> -----
> Frames	Periods	F*P/SR	Measured
> 
> 64	2	2.9	8.2
> 128	2	5.8	11.1
> 256	2	11.6	15.9
> 512	2	23.2	28.5
> 1024	2	46.4	52.7
> 2048	2	92.9	102.1
> 4096	2	185.8	187.1
> 
> 64	3	4.4	9.7
> 128	3	8.7	14.0
> 256	3	17.4	22.7
> 512	3	34.8	30.1
> 1024	3	69.7	65.0
> 2048	3	139.3	138.6
> 4096	3	278.6	279.9
> 
> 
> SR 48000
> -----
> Frames	Periods	F*P/SR	Measured
> 
> 64	2	2.7	7.9
> 128	2	5.3	10.5
> 256	2	10.7	15.9
> 512	2	21.3	26.5
> 1024	2	42.7	48.9
> 2048	2	85.3	94.5
> 4096	2	170.7	171.9
> 
> 64	3	4	9.2
> 128	3	8	13.2
> 256	3	16	21.2
> 512	3	32	27.2
> 1024	3	64	60.2
> 2048	3	128	127.2
> 4096	3	256	257.2
> 
> 
> Questions to the wise guys:
> --------------------------
> 
> 1)
> As probably everyone has noticed, with 3 periods/buffer and
> frames/period of 512, 1024 and 2048 the measured latency is smaller
> than the calculated latency. I don't see how this can be possible
> since, as far as I understand it, the measured latency should be the
> calculated + some.
> 
> 
> 2)
> When doing those measurements I noticed that with 44100 the Frames
> reported by jdelay always ended in .732 Frames and with 48000 always
> ended in .761 Frames.
> 
> I found out that I can change that value by changing the sensitivity
> of the input or the amplification of the output. In the case the
> variance is below one frame, but I find it very strange that it's
> always the same at a given amplification/sensitivity.
> 
> 
> 3)
> Is it normal that the difference between calculated and measured
> latency varies so much? Shouldn't it just always be the same?
> 
> 
> 4)
> Should these results be similar with any usb interface?
> What does have most influence on such results?
> 
> 
> I hope this can be of use after the remaining questions have been
> answered.
> 
> Best regards,
> Philipp


Don't rely on those results.
I just had very different results for the same setting.

48kHz, 256/3 had 21.2ms, now I also got 15.9ms and 20.2ms, and I have
no idea what caused that.
Note that it never occurred while jdelay was running but have a look at
this:

1017.757 frames
^C
hollunder ~  $  jdelay
761.754 frames


I absolutely don't know how this can be, what happened in these
1-2 seconds in between, what's at fault, and it looks like it could be a
serious issue.

Regards
Philipp