Re: [linux-audio-dev] MuCoS: What's going on?

New Message Reply About this list Date view Thread view Subject view Author view Other groups

Subject: Re: [linux-audio-dev] MuCoS: What's going on?
From: Stefan Westerfeld (stefan_AT_space.twc.de)
Date: pe tammi  07 2000 - 12:41:43 EST


   Hi!

On Thu, Jan 06, 2000 at 01:17:29PM +0100, Benno Senoner wrote:
> On Wed, 05 Jan 2000, David Olofson wrote:
> > Hi!
> >
> > Still alive, although it doesn't show much on the MuCoS site... :-(
> > Time to get working again!
> >
> > What's going on with the related project and prototypes? (Benno,
> > Paul,...?)
>
> Some time ago I began to experiment with a client/server low-latency
> buit ran into troubles (too high latencies, semaphore deadlocks etc).
>
> I am now working on a simple userspace latency-profiling API, which
> should give you latency-graphs similar to these produced by "latencytest".
>
> This should help us to better identify where the bottlenecks in
> low-latency code are, especially try to optimize the client/server API.
>
> The kernel may require further little refinements (semaphore code etc),
> but msgrcv() and msgsnd() seems already promising.
>
> [...]
>
> I hope that a preliminary latency API is ready in a few days, then I can begin
> to profile the IPC code of the client/server.

Finding an optimal IPC mechanism is really a good idea. I've currently
implemented aRts/MCOP communication on top of TCP streams, which gives a
performance of 4500 synchronous invocations per second (that is: client
makes a call, server does something, client receives the result).

While this is factor three faster against the old mico solution for
instance, it is still way too slow. The bottleneck really seems to be
the

- client: writing data via TCP
- client: dropping to select
- server: waking up from select
- server: reading data from TCP
- server: writing result to TCP
- server: dropping to select
- client: waking up from select
- client: reading from TCP

steps, which probably consume most of the time (marshalling, demarshalling
and dispatching are really fast).

I also don't know how much the process context switching costs, but I hope
things like 30000 invocations could be possible.

So I'm really looking forward to your results ;)

   Cu... Stefan

-- 
  -* Stefan Westerfeld, stefan_AT_space.twc.de (PGP!), Hamburg/Germany
     KDE Developer, project infos at http://space.twc.de/~stefan/kde *-


New Message Reply About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b28 : pe maalis 10 2000 - 07:23:26 EST