KLANG: A New Linux Audio System For The Kernel

[Staffan]

"professional grade audio, that means lowest possible latency, latency compensation and bit exact precision at a very low CPU load. KLANG has been designed as a signal routing system, supporting seamless and transparent signal transport between all endpoints.

Isn't this what ALL the other sound systems have been trying to do? They have all failed.

ALSA has worked fine for me for as long as I can remember. I don't want another square wheel.

[Awesomeness]

Let me remind you principle that ANY engineer follows, IF IT'S AINT BROKEN - DON'T FIX IT!

Funny, considering that according to a post of the KLANG author, ALSA is currently broken from an engineering point of view.
Read it if you can (it’s German, maybe some translation tool helps): http://www.heise.de/open/news/foren/...22197656/read/

[PeterKraus]

I am an ALSA user, and while I don't see anything wrong with it, I see what this project is trying to address - putting hardware management back into the kernel. The same thing happened with video (KMS), and people were initially complaining, but now everyone is happier. Furthermore, if it is going to be transparent to OSS-supporting apps, it will be a drop-in replacement. I mean, ALSA supports OSS-compatible apps, Pulse supports ALSA and OSS... This is more of a "reworking the plumbing" than "reworking the dashboard".

[Neco-Link]

Why not just port Alsa or PA into the Kernel?
Alsa and PA are working really fine or at least for me.
And it seems OSS4 ain't really popular here.
So what are the technical implications and why isn't he doing it?

[89c51]

http://ardour.org/pd_on_klang

Paul could you also answer why mac os x does a better job with a single "stack", why we can't have the same in linux and also why there is no effort trying to merge jack and PA (lack of manpower was the reason last time i heard about it).

[ninez]

http://ardour.org/pd_on_klang

Wow, Paul. i didn't even realize you visited/post @ phoronix.

Thanks for posting this information. I was aware of the floating point problem, as well as re-writing drivers problem, but others i had not really thought about too in depth. it's also nice to hear the perspective of someone who is in the know, and as knowledgeable about this stuff as you are. thanks

But, regardless of reading your post @ ardour.org ~ i think KLANG is generally a stupid idea. We don't need yet another sound-system for linux. if there are problems with ALSA, PA and Jack - it seems for more reasonable and realistic to work on those instead.

I did also find Ben l0ftis (from harrison consoles) comments interesting;

I'd like to reiterate a few points based on our experience with Linux audio at Harrison.

We have a product called Xdubber that was developed 5 years ago. The Xdubber uses a custom JACK driver (no ALSA). The system operates with 8-sample buffers ( compared to the much more common 1024 or, at best, 64 samples provided in most systems ) for extremely low latency. We send 64 channels of 96kHz audio in and out of the system. I have tested this system for days using an Audio Precision bit-test with error-checking turned on.

Our findings with an actual commercial product have shown that:
*JACK has a very minimal CPU/memory footprint.
*JACK has nothing to do with XRUNS. it reports xruns that happen in the driver and/or application level which otherwise would go unreported.
*There is no fundamental issue with the coding style of JACK. It's just very hard work to d this kind of plumbing.
*An ultra-high performance, ultra-low latency system can be built with JACK

I've had similarly good experiences with the best-implemented ALSA devices, such as RME and older M-Audio.

There ARE a lot of issues with linux audio. But they stem mostly from the unbelievably wide range of use-cases between applications and devices. And the fact that Linux users actually have higher expectations of their audio system .... for example Windows still doesn't have the concept of virtual MIDI ports, much less inter-application routing. OSX's CoreAudio is better but still lacks fundamental features of ALSA and JACK.

-Ben Loftis
Harrison Consoles

i thought i'd post his comments for those who are lazy, or missed Paul's original post.

[kokoko3k]

The OS X kernel has RT facilities. But you're missing the point: you don't need RT for this. The reason why RT is used in Linux for audio just shows the problems the audio stack has. RT is not needed, unless you're doing something wrong in the audio infrastructure.

Given the fact that the audio subsystem works into userspace,
if you need a latency around 5ms when the kernel has to process tons of events (interrupts), then you need real time responces for the audio thread and the kernel needs to be preemptible.

Standard linux can't do it right now, even by tweaking granularity settings:
http://doc.opensuse.org/documentatio...scheduler.html
...and/or givin a process a real time priority RR/FIFO

So if the video driver at kernel level is interrupt driven, and block it hardly during an effect, the audio skips is perfectly normal.

Building a preemptible kernel via rt patchsets or having the audio subsystem at kernel level will work.

[9a3eedi]

Normally, I'd be interested, but now that we have all these different Linux audio standards this generally feels pointless. The guy does have good points of wanting to put all the audio stuff into the kernel, but is it really that necessary?

I'm no audio engineer or musician, but isn't PulseAudio doing a fairly good job as far as low latency goes? It had problems when it was first introduced, but not so much anymore from what I understand, and generally many people just seem satisfied with it. It should be good enough for games at least, right?

And audio engineers, like many others mentioned here, can simply use JACK and a low latency kernel if they need that low of a latency for their stuff.

[cb88]

You know...PulseAudio I despise the extra daemon running in the background. Oh.. *Idea* lets hide the whole shebang in the kernel! that way they can't see it right off... You know it really would be better if the parts that have to be running all the time got merged into the kernel and the parts that don't stayed in userspace. The problem with this is audio problems are now harder to fix as it increases the likelihood of requiring a kernel recompile.

Clearly you can use FP in the kernel http://www.linuxsmiths.com/blog/?p=253 I can see how there could be drawbacks however if you tried doing alot of FP in the kernel.

[Ferdinand]

Even if his chances of seeing the community accept broadly yet another sound system are thin IMHO, if this guy could manage to create a sound system that is

1) simple in its architecture (no self-sentient userspace daemon required);
2) simple in its usage (zero configuration required at least to put the audio hardware in a sane state);
3) moderate in its requirements (a pentium 4 should be able to play an mp3 without suffering at all);

then it might be a good thing.

Shouldn't a pentium 133Mhz be able to play an mp3 fluently?