too many wakeups/sec?

[Jussi Kukkonen-3]

Hi,

I would have filed this as a bug but bugs.xiph.org doesn't seem to work
(despite quite enthusiastically claiming to :)). So here goes:

I'm seeing quite a few wakeups/sec in icecast2 2.3.2 on Debian Testing.
In fact, whether icecast is doing anything or not, powertop reports 44.3
wakeups per second. This is pretty big number, on my laptop it means
icecast alone is responsible for every third wakeup on idle.

I have no time to start digging through the code right now, but please
post if you have a hunch where these wakeups are coming from. If I have
time later, I'll  consult this thread.

  - Jussi
_______________________________________________
Icecast-dev mailing list
Icecast-dev@...
http://lists.xiph.org/mailman/listinfo/icecast-dev

[Karl Heyes]

On 21/10/09 10:56, Jussi Kukkonen wrote:

> Hi,
>
> I would have filed this as a bug but bugs.xiph.org doesn't seem to work
> (despite quite enthusiastically claiming to :)). So here goes:
>
> I'm seeing quite a few wakeups/sec in icecast2 2.3.2 on Debian Testing.
> In fact, whether icecast is doing anything or not, powertop reports 44.3
> wakeups per second. This is pretty big number, on my laptop it means
> icecast alone is responsible for every third wakeup on idle.
>
> I have no time to start digging through the code right now, but please
> post if you have a hunch where these wakeups are coming from. If I have
> time later, I'll  consult this thread.

This will vary depending on what you have icecast doing but you should
find that the trunk code has less wakeups.  The connection and stats
parts are not quite the same as the kh tree, although they have been
changed, so you should see an improvement.

karl.


_______________________________________________
Icecast-dev mailing list
Icecast-dev@...
http://lists.xiph.org/mailman/listinfo/icecast-dev

[Jussi Kukkonen-3]

On Wed, Oct 21, 2009 at 5:53 PM, Karl Heyes <karl@...> wrote:
>> I'm seeing quite a few wakeups/sec in icecast2 2.3.2 on Debian Testing.
>> In fact, whether icecast is doing anything or not, powertop reports 44.3
>> wakeups per second. This is pretty big number, on my laptop it means
>> icecast alone is responsible for every third wakeup on idle.
>
> This will vary depending on what you have icecast doing but you should find
> that the trunk code has less wakeups.  The connection and stats parts are
> not quite the same as the kh tree, although they have been changed, so you
> should see an improvement.

Thanks, Karl. Testing with trunk I can confirm this. Just as before it
doesn't seem to make difference if icecast streams something or not,
but there is a significant improvement compared to 2.3.2:

2.3.2     44 wakeups/s
trunk     13 wakeups/s

13 wakeups is still not optimal when in reality my icecast usually has
nothing to do (no mounts, no clients), but it is a lot better.

Thanks for the good work,
  - Jussi
_______________________________________________
Icecast-dev mailing list
Icecast-dev@...
http://lists.xiph.org/mailman/listinfo/icecast-dev

[Jussi Kukkonen-3]

Karl asked me to test with his kh17 branch and I did.

It looks like this is the first time there is a difference between not
doing anything and actually streaming (this could be a result of my
very non-scientific testing but the numbers looked fairly constant).

version     wakeups/s when       wakeups/s with
           idle (no mounts)     1 mount, 1 stream
-------------------------------------------
2.3.2       44                   44
trunk       13                   13
kh17         4                   10


This is a lot better for my use case case (mostly idle). Even the
difference when streaming looks significant: the number of wakeups
stays fairly stable over time so the 13 vs 10 difference should be a
statistically significant one.

- Jussi
_______________________________________________
Icecast-dev mailing list
Icecast-dev@...
http://lists.xiph.org/mailman/listinfo/icecast-dev

[Jussi Kukkonen-3]

...and one more test with kh17b. From 2.3.2 to kh17b the "wakeups/s"
figure has come down 84% - 98% in my test cases. I have very little
idea what Karl's been doing but from a user POV the results look very
good.

version     wakeups/s when       wakeups/s with
            idle (no mounts)     1 mount, 1 stream
-------------------------------------------
2.3.2       44                   44
trunk       13                   13
kh17         4                   10
kh17b        1                    7

 - Jussi
_______________________________________________
Icecast-dev mailing list
Icecast-dev@...
http://lists.xiph.org/mailman/listinfo/icecast-dev