Re: [linux-audio-user] Re: fst-1.8 triggers wine-preloader processes using a lot of ram

From: Lee Revell <rlrevell@email-addr-hidden-job.com>
Date: Sat Jun 24 2006 - 00:00:06 EEST

On Fri, 2006-06-23 at 12:11 +0200, Benjamin Fabricius wrote:
> Am Donnerstag, den 22.06.2006, 22:14 -0700 schrieb Kjetil S. Matheussen:
>
> > not totally sure what you're doing, but keep in mind that Muse put a >
> > lot
> > of effort into making this work reliably across many plugins. don't
> > expect to see this happen "easily".
>
> what i want and am doing is to route a maximum of 64 audio channels
> through various plugins by using jacks routing matrix and then output
> these again. the plugins are being loaded by fst. ive ran some latency
> and signal tests on this system to check if the system is stable and bit
> transparent and also estimated s/n, distortion factor etc and everything
> is working to my satisfaction thanks to having built this system using a
> more or less embedded linux. this seems to offer enough cpu ressources
> for this setup to work (dedicated and thus) reliably with up to 18 (max.
> possible plugs to this date) plugins without getting xruns or other
> artifacts. so thats what im doing. yesterday while testing suddenly
> qjackctl crashes.. due to an out of memory fault. ive got
> 512MB ram minus the space for my ramdisk. a quick look into top now saw
> me confronted with a large list of wine-preloader processes using up to
> 10% of my memory each (!).
>
> i was not doubting the effort invested in making plugins run stable with
> fst or Muse etc. i merely thought it was worth finding out if any one
> has had these difficulties in an environment with similar low
> ressources.
>
> as a matter of fact i would actually like to stick with fst because i am
> surprised it works as reliably as i have experienced.

Try lowering RLIMIT_STACK from 8MB to 256KB or something (the default is
set in /etc/security/limits.conf). Heavily mulithreaded processes +
mlockall() wastes a LOT of memory for thread stacks.

The latest SVN Jack solves this problem by setting the thread stack size
itself.

Lee
Received on Sat Jun 24 04:15:08 2006

This archive was generated by hypermail 2.1.8 : Sat Jun 24 2006 - 04:15:08 EEST