Subject: Re: [linux-audio-dev] LAAGA API Proposal
From: Richard Guenther (rguenth_AT_tat.physik.uni-tuebingen.de)
Date: Fri Jun 15 2001 - 11:14:41 EEST
On Thu, 14 Jun 2001, Vincent Touquet wrote:
> Maybe we could nail the synchronous vs. asynchronous discussion down to
> this:
>
> Synchronous (blocking) programming techniques work perfectly for low to
> medium loads and has the _lowest_ latency.
Yes. It gets lowest possible latency. But for LAAGA where you pay the
context switch costs the difference is negligible (but may be noticable).
> Asynchronous (non-blocking) programming techniques continue directly
> after some calls and are notified when requested data is available /
> more data can be sent (as a cause of that call) and then they have to
> react. This _increases_ latency, but it makes the program also work
> under abusively high loads, something the blocking programming style
> will not be able to cope with.
Yes - especially if you have computing power (more processors) unused
in the sync. case.
> So I think if we want *the lowest* latency, the way to go is blocking io ...
> When we are aiming for <3 ms latencies, 5 ms latency penalties are not
> acceptable.
>
> It is a matter of priorities I gues ;)
Sure :) For the proposed setup of LAAGA apps (<50 apps, <~1000 connections
- probably way lower usually) this is even more a matter of taste than
a matter of latency. For LADSPA networks (inside one app, w/o context
switches and lots of nodes, >100) sync. operation makes very much sense
and I can see GLAME having much lower latency in the worst case than
such a LADSPA network. But these facts do not stand for LAAGA - oh, did
I mention I dont want to encourage people to use LAAGA in an "inside
app" mode?
Richard.
> Regards
> Vincent
>
> PS: I base myself on
> http://www.sysadminmag.com/articles/2001/0107/0107a/0107a.htm ...
> for this short reaction. (credit where credit is due + you get a nice
> benchmark extra)
>
>
-- Richard Guenther <richard.guenther_AT_uni-tuebingen.de> WWW: http://www.tat.physik.uni-tuebingen.de/~rguenth/ The GLAME Project: http://www.glame.de/
This archive was generated by hypermail 2b28 : Fri Jun 15 2001 - 13:12:59 EEST