[LAU] Latency and USB interfaces

[Clemens Ladisch]

Tim Goetze wrote:
> [Clemens Ladisch]
>> In most cases, the ALSA buffer/period parameters directly reflect
>> hardware parameters, but in the case of protocols like USB where it is
>> not possible to transfer data directly out of a ring buffer, there is an
>> additional packet queue, which adds latency.  (The length of that queue
>> is either constant or proportional to the period size, depending on the
>> driver.)
>
> Thanks you Clemens for clearing that up.  How would one go about
> determining the size of this queue between ALSA and USB?

In the case of the current version of snd-usb-audio, the queue has the
same length as the buffer, but never more than 18 ms.

> And is it correct to assume that in a best-case scenario where the
> sample clock is synchronised to the USB clock

Pro devices typically don't do this.

> and the period size coincides with the interval at which USB frames
> are signalled,

This does not happen with Jack, where period sizes typically are a power
of two.

> the additional latency will be one period?

These two conditions have nothing to do with the queue length; they
ensure that USB interrupts happen exactly at period boundaries, i.e.,
that there is no jitter in the reported period interrupts.


Regards,
Clemens