[haiku-commits] Re: haiku: hrev50603 - src/kits/media headers/os/media

  • From: Axel Dörfler <axeld@xxxxxxxxxxxxxxxx>
  • To: haiku-commits@xxxxxxxxxxxxx
  • Date: Thu, 20 Oct 2016 13:55:36 +0200

Am 20/10/2016 um 13:50 schrieb Dario Casalinuovo:

At the time my test was to add a "snooze(100)" in different parts of the
chain depending on the situation I wanted to simulate :
[...]
Trying to give a more meaningful value results always in ever increasing
latency or cracking sound.

And that no longer happens now?

> And how can you compute lateness without enqueue time?
I'm not saying the enqueue_time doesn't have it's meaning but alone
isn't going to solve anything. As said various times, perhaps it could
be solved by changing completely how latency is handled through nodes,
having a thread activation log and looking at the whole chain behavior
once every buffer reach the final consumer. In any case I might be wrong
so feel free to adjust my errors.

Well, you need the enqueue time in order to correctly identify the node which is responsible for the lateness.
IOW now the latency might be attributed to a node higher in the chain, eventually causing a longer latency for the whole system.
However, what was done with the enqueue time didn't look exactly useful, either, so in that sense I can understand why it didn't improve anything.

Bye,
   Axel.


Other related posts: