Subject: Re: [linux-audio-user] Advice needed re: latency tuning/optimization, esp. video
From: Chris Metzler (cmetzler_AT_speakeasy.net)
Date: Tue Feb 17 2004 - 22:56:26 EET
On Mon, 16 Feb 2004 20:15:31 +0100
Robert Jonsson <robert.jonsson_AT_dataductus.se> wrote:
>
> Hi,
Hi. Thanks for your reply!
>> The sound card shares IRQ 10 with the USB2 bus; but I have no USB
>> devices of any sort. On IRQ 9 are two of my four IDE channels.
>
> This is probably not optimal, if I understand correctly irq10 is a bit
> down in the priority list, if you haven't read the lowlatency howto, do
> that.
I have read it, if you're referring to:
http://www.djcj.org/LAU/guide/Low_latency-Mini-HOWTO.php3
and I've spent a fair amount of time googling around for other low-
latency suggestions/comments, such as the ones on the LAD site.
My understanding is that the only higher-priority IRQs than IRQ10
are 0, 1, 8 and 9; and 0, 1 and 8 are taken. I know that IRQ10 isn't
optimal for the soundcard -- that IRQ9 would be better -- but I didn't
want to start screwing around with IRQ assignment until I was sure I
needed to, because it's something I'm still fairly ignorant about. My
BIOS does allow me to assign IRQs to PCI slots -- so I could force the
soundcard onto IRQ09 -- but I don't seem to be able to monkey with the
IRQs assigned to the four IDE channels, or AGP, or USB or IEEE-1394.
Fortunately I'm not using any USB or Firewire devices at all, so I
don't expect interrupts to be generated for them; but certainly for
the disks and video card there will be.
But despite the fact that two of my four IDE controllers sit on the
one IRQ with higher priority than the soundcard, heavy disk use doesn't
seem to cause me that much trouble. At least, it didn't in the
low-latency-png test I tried (mentioned in the previous email). It
was video that caused the trouble, and video sits on IRQ11, of lower
priority than the soundcard on IRQ10.
09 -- IDE2, IDE3, USB1
10 -- IEEE-1394, USB2, SBLive
11 -- USB1, AGP
12 -- mouse
13 --
14 -- IDE0
15 -- IDE1
03 -- ttyS01
04 -- ttyS00
05 -- NIC
06 --
07 -- parallel printer
><...>
>> This result surprises me. I'm using a video card that's supposedly
>> very very good at 2D stuff -- indeed, it's the video card that RME
>> recommended as recently as last year as their card of choice for
>> audio workstations. It's definitely open at AGP 4x (XF86 wants to
>> open it at 1x unless you explicitly say in the config file that you
>> want 4x). 32 MB isn't a huge amount of video RAM, but I would think
>> would be fine for 2D stuff.
>
> That a card is good for graphics doesn't necessarily mean that it's good
> for audio. That RME recommends it points to that the card should be well
> behaving hardware wise. Things to check:
> - That it has an irq that has lower priority than the audio card ...
> might not be possible with agp ?
It does.
> - Try and change the AGP setting. Faster isn't necessarily better.
I will. I woulda naively figured that (among other things) faster
meant "releases the kernel's attention more speedily" or something
like that. But maybe not. Maybe it also means "asks for stuff more
frequently" too. We'll see.
> The most probable offender is the driver though, I don't know if there
> are multiple drivers for this card so you could try another?
I didn't think about this at all, and it's definitely a possibility
now that I do think about it.
I'm using XF86 4.2 rather than 4.3 because I'm a Debianista running
"sarge", the release that's being assembled right now and is targeted
for this summer. X 4.3 hasn't made it into Debian yet, even in its
"unstable" distribution. This matters because the Matrox driver in
XF86 4.2 doesn't explicitly support my video card. It *does* support
the G400, the immediately previous version; but not the G550. The
G550 is supported in XF86 4.3. I wouldn't have immediately thought
of this because I've never run into an issue with 2D before; the
driver I've used has done fine. The only problems I've had in the
past has been with 3D/DRI/OpenGL stuff -- odd lockups running
FlightGear, for instance. But for 2D it's been fine. But now I'm
wondering.
Matrox themselves provide (open-sourced!) drivers for their cards;
but they're basically the same code as the XF86 versions. At one
point, trying to run down the 3D issue, I switched to theirs for a
while. It didn't help.
Anyway, this is definitely something to think about, thanks muchly.
But other than try to backport XF86 4.3, I'm not sure what I can
do in response . . .
> If you have a graphic login (e.g. you login directly to X) you should
> check that xdm/kdm or whatever display manager you are running does not
> have elevated priority. Mandrake does this as has been noted here
> before, for mandrake it's in:
> /etc/X11/xdm/Xservers
> the line is
> :0 local /bin/nice -n -10 /usr/X11R6/bin/X -deferglyphs 16
> -10 is wrong 0 is _more_ right.
I'm OK here.
> If you are running KDE or GNOME there might be issues with certain
> functions of the desktop environment. I'm running KDE myself with no
> problems but others have reported problems with both.
I'm usually under GNOME, and I've worried about that a little for when
I get to doing more serious stuff. But in the meantime, the X11 stress
test done in the low-latency test suite seems oblivious to desktop
environment. It grabs a chunk of screen in the corner and draws on it
in a way that looks very libXt and nothing else.
> I guess you could try different graphical settings also, amount of
> colors, display size... no solutions but it might help to identify the
> perpetrator.
>
> All I can think of right now. Hopefully someone with a similar
> card/setup can comment with better ideas.
No, these were good. Thanks.
-c
P.S. cc'd to make sure you'd see the response, because I couldn't
respond for a day. Sorry/let me know if CCs annoy.
-- Chris Metzler cmetzler_AT_speakeasy.snip-me.net (remove "snip-me." to email)"As a child I understood how to give; I have forgotten this grace since I have become civilized." - Chief Luther Standing Bear
This archive was generated by hypermail 2b28 : Tue Feb 17 2004 - 23:01:20 EET