[sdiy] bootable linux cd
Magnus Danielson
cfmd at swipnet.se
Mon Jul 28 00:12:43 CEST 2003
From: Rainer Buchty <buchty at cs.tum.edu>
Subject: Re: [sdiy] bootable linux cd
Date: Sun, 27 Jul 2003 23:45:05 +0200 (CEST)
> > Sorry, but writing a new OS, even if dedicated, is not what I think is
> > the best solution if you go beyond a certain level.
>
> You're preaching to the choir... Personally, I would go with Linux as a
> start and apply RT patches and/or package timing-critical stuff into
> kernel modules.
Actually, if you look in at the 2.5.x stuff, going for 2.6.x, you will find
quite alot of usefull changes is already there. There is still a few patches
you could apply, or you go fullblown RT. However, if you look into the
infrastructure allready in place (Low latency patches, ALSA, JACK, LADSPA, etc,
etc.) I think almost all the things needed is there. Then, bring on the big
knife and cut away all the things which doesn't need to be there... I think
that you have a very nice starting-point doing that.
It isn't as easy as it used to be to dismiss Linux as a solution for sound.
This is why it is creeping into more and more applications where dedicated OSes
used to be the only way. It is becomming a rationalization of an industry, and
to some degree happening in the quite. Quite many Set-top boxes run Linux these
days for instance. For them it is really cutting time and they can concentrate
their efforts on shaping a product.
Experiences with certain commercial RTOSes have shown that many of them is way
behind for many applications anyway, and the way you work with a modern OS in
form of development tools either does not exists or doesn't come cheap. There
is increasingly more nifty tools popping up for debugging and tracing which
makes it easier and easier to actually get to know what is happening.
> My posting was rather "instead of going with DOS as a base OS, you
> might consider just writing your own MusicOS from scratch" than a
> commitment to "YAOS".
Well, you can... but on the same time one must be aware that we've gotten
increasingly spoiled, and there is a good reason for that, and that is that the
stuff we do today is so much more complex that we can't really do usefull work
with tools designed for problems being a fraction of what we do today. But, if
you still is doing designs of a certain size, then working with tools for that
size is still a good thing. If you have a very simple thing which may be a few
hundred lines of assembler, then that is still a valid tool. If you have a more
advanced CPU with paging, caches, etc. etc. there is a number of additional
concerns. If you toss on real time requirements when also asynchronous events
can come in and consume time you start to get messy. Add to that making sure
various routines get good time-properties, you avoid locks, that the OS can
keep multiple events in the track in the same time, while avoiding trashing the
cache... is maybe not my definition of joyfull musical programming. However, if
you want to learn, then please go ahead!
There is a few good books from the Linux kernel world that I would recommend
reading anyway, since many problems will not go away even if you send the code
out the door.
I'm not very good at kernel work, but I have some insight into some of the
problems and I've seen problems in real life. I've given a driver expert a hard
time by asking him to hack the bloody driver until the scope views the
waveforms right rather than claiming that my design behaved incorrectly. Let's
just say that 8kHz rock steady interrupt requests put requirements on older
kernels as well as new. Making bad decissions will make things bubble around
like you never could imagine. Interrupts is a beast sometimes.
> > If you want to do a music app, you want to concentrate on doing that
> > well rather than concentrating on doing an OS well.
>
> Fully agreed.
Right.
Now I am going back to my dungeon to see why ALSA doesn't find my soundcard so
that JACKD can run with ALSA so that I can run Ardour which I want to test out.
I didn't say it wasn't a road without problems, but usually I can figure things
out a hell a lot more quicker in this OS than any others... so I don't complain
too much.
Cheers,
Magnus
More information about the Synth-diy
mailing list