Subject: Re: [linux-audio-dev] Newby(t)e
From: Benno Senoner (sbenno_AT_em.gardena.net)
Date: ke marras 24 1999 - 09:15:22 EST
>
> Benno Senoner <sbenno_AT_gardena.net>
> * Client/server implementation and API.
> * Various performance testing tools.
> * Some demonstration/proof-of-concept programs.
>
I got a client/server implementation hack running,
using sharedmem +SYS V sempaphores:
(semop)
but still many problems:
on my celeron box, when running on an idle system,
I get no droputs using a 2.8ms audio buffer with server
and clients exchanging 128bytes at time.
But when I put some load on the box there are audible dropouts.
Using 2 clients the semaphores deadlock after a while with the server waiting
for the clients and the clients waiting for servers.
I will try to do some tests using threads instead of processes
and pthread_mutex instead of semaphores, lewt's see if this
gets better.
But as far I know there should not be much
difference since threads and processes are almost the same on
linux.
When I will get all the testing stuff togheter I will do
a detailed explanation of the test.
I'm not sure if this semaphore deadlock is really my
own fault (programming mistake) or a kernel bug ...
we will see. (I will CC linux-kernel as well ..)
As for the latency in semaphore code I'm confident that Ingo will
find the place where to add the scheduler hookup,
in order to get down to our 2ms standard.
Benno.
This archive was generated by hypermail 2b28 : pe maalis 10 2000 - 07:23:25 EST