From alsa-devel-owner@alsa.jcu.cz  Thu Feb  4 13:32:37 1999
Received: from keg.cs.vu.nl (root@keg.cs.vu.nl [130.37.24.4])
	by marvin.jcu.cz (8.9.1a/8.9.1) with ESMTP id NAA17367
	for <alsa-devel@alsa.jcu.cz>; Thu, 4 Feb 1999 13:32:19 +0100
Received: from localhost by keg.cs.vu.nl with esmtp
	(Smail3.1.28.1 #49) id m108NxA-0007ESC; Thu, 4 Feb 99 13:32 +0100
Date: Thu, 4 Feb 1999 13:32:16 +0100 (MET)
From: Andy Lo A Foe <arloafoe@cs.vu.nl>
To: alsa-devel@alsa.jcu.cz
cc: R Pickett <emerson@hayseed.net>
Subject: Re: Writing Article Re: Linux audio;  also have new hardware support
In-Reply-To: <Pine.LNX.3.96.990204103122.8890D-100000@entry.jcu.cz>
Message-ID: <Pine.GSO.4.05.9902041316440.27444-100000@keg.cs.vu.nl>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Reply-To: alsa-devel@alsa.jcu.cz
Sender: alsa-devel-owner@alsa.jcu.cz
Precedence: list

On Thu, 4 Feb 1999, Jaroslav Kysela wrote:

> > That's good.  For the pro market, latency is a BIG issue.  Currently, BeOS is
> > on everyone's mind in the MI industry because it has < 5ms latency guaranteed.
> > In order for Linux to be a compelling pro audio platform, it at least needs to
> > approach, if not exceed, this performance.
> 
> I don't expect very much improvement. We can currently with good hardware
> (PCI soundcards) reach 1.5ms latency.

I just did some tests. It looks like 2.2.1 is a lot worse at real-time
scheduling than previous versions. When doing playback with small fragment
sizes and number (512 byte fragments with 4 fragments total) simply
reading some data from disk almost always causes underruns (this is with the soundcard
feeding process running with SCHED_FIFO and the max priority. When I do a
dd if=file.mp3 of=/dev/null before starting to play (so the whole file is
in buffer cache) all the underruns go away. This lets me conclude that, at
least, disk access is interfering with scheduling. It also doesn't matter
wether you're reading from SCSI or EIDE disks, same horrible results...

Running latency.c in alsa-lib/test seems to confirm somethings wrong
with 2.2.1 and ALSA (on my machine at least). Can anyone else please
run the latency program with kernel 2.2.1 and ALSA CVS? I get an average 
optimium fragment size of 1216, which is unbelievably bad :(

Thanks..

Andy
--
Andy Lo A Foe, arloafoe@cs.vu.nl |   Linux v2.2.1   | PII 266 
http://www.cs.vu.nl/~arloafoe    | GSM: 0622-440695 | Bebox 133
http://orbital.xs4all.nl/andy    |   ICQ: 7472949   | IRCNet: Adnans


