[pulseaudio-discuss] [PATCH] Do not use tsched watermark if tsched is disabled

[David Henningsson]

On 2010-09-13 13:03, Colin Guthrie wrote:
> 'Twas brillig, and David Henningsson at 13/09/10 11:14 did gyre and gimble:
>> On 2010-09-04 14:10, Colin Guthrie wrote:
>>> I'd be interested as to whether anyone else can repeat this experiment
>>> and get similar results. Do you guys get a broken chordtest too (it's on
>>> the RedHat bug I mentioned at the beginning of this thread)?
>>
>> I have now tried to repeat the experiment. The chordtest.sh seems to be
>> buggy in itself (the cleanup does not remove the gst-launch, which in
>> turn had to be renamed to gst-launch-0.10 here).
>
> Yeah I have gst-launch-0.10 here too... not quite sure why, I'd have
> thought we could ditch the old 0.8 support by now but hey ho. (I don't
> follow gst dev super closely)
>
> I thought that the script trapped ctrl+c and killed any processes
> started. It seems to be clean for me.
>
> Perhaps the problem is that /bin/sh is not actually bash on your system?

It is "dash" on Ubuntu systems.

> Perhaps just changing the first line to:
>
> #!/bin/bash would cause it to tidy things up properly?

Yes, that worked, thanks.

>> Anyway, the results
>> were not encouraging - with tsched=0, pavucontrol, and -vvvv to syslog
>> on, three tones were heard, then things went quiet - however, pulseaudio
>> started to eat more and more memory. Quickly my machine started swapping
>> and became unresponsive, so I killed PA.
>> Besides that, when I looked at pavucontrol, only the meters of the first
>> three were moving, the other ones were silent. My log got filled up with
>> "memblockq: pool full" as well. I'm getting the feeling that this
>> problem is something different, unrelated to DMA controller hardware.
>
> Interesting, can't say I noticed this, but I probably wasn't looking
> closely enough.
>
>> My suggestion is that you should commit your proposed patch as it
>> improves the situation compared to the current situation. If there are
>> additional problems, let's nail them down separately.
>
> OK, sounds reasonable. Do you think the patch I posted is OK with the
> 1330 time?

I think it is good enough for now. If it turns out to be too little, we 
can adjust it later.

> I guess it's not super important as if it solves your original problem
> that kicked off this whole thread, then that's the main thing!!

That PA can handle a stress test is important, but it's a different issue.

-- 
David Henningsson, Canonical Ltd.
http://launchpad.net/~diwic