Linked by Filippo Pappalardo on Wed 14th Apr 2004 07:36 UTC
Multimedia, AV I wanted to write something about the great progress being carried on linux as OS of choice for a professional Digital Audio Workstation (DAW) since a long time. With the inclusion of the Advanced Linux Sound Architecture (ALSA) into the 2.6 kernels, time has come to extend my experiences to all of you.
Permalink for comment
To read all comments associated with this story, please click here.
Various comments.
by Chris Metzler on Wed 14th Apr 2004 16:42 UTC

Regarding the suitability of Linux for pro-audio work . . .

At present, using JACK and a preemptable/low-latency kernel (I'm using a patched 2.4; I don't have personal experience with 2.6, but it's supposedly an improvement), the maximum latencies I've seen have been at about the 2ms level, with sub-millisecond latency the norm. The kernel buys you the low-latency; JACK ensures that user-space delays don't ruin it, and further makes that latency homogenous through appropriate buffering of signals. In general, Linux scoops Windows on this end.

Where Windows scoops Linux, of course, is in the applications -- the level of features and their robustness in areas such as HDRs, sequencers, samplers, etc. But anyone who's been following this closely for a while knows that the situation has been changing dramatically over the last few years. Three years ago, a comparison between Rosegarden ( ) and Cubase would have been absurd. Now, it's not absurd. Cubase still wins, but the gap is much narrower. For most work and most people, Rosegarden is just fine; and I question whether Cubase will still be a better product in two years.
Similarly, Ardour ( ) is intended to fill the role normally occupied by ProTools. Does it succeed? Not completely; but it's pretty damned good, and it's getting better a lot faster than ProTools is. There are lots and lots of Linux audio apps ( ), which range in completeness and quality from nearly unusable to excellent.

In my opinion, pro-audio on Linux is a comer. But that doesn't mean there aren't issues to solve.

One big one is a common bugaboo for open source projects -- documentation. The documentation for most of these projects is awful. This is especially important for applications like this, which are farther removed from the kind of thing people usually do on computers. If there wasn't any user-oriented documentation for GNOME or KDE, I think I could still figure things out just fine; but if something more specific like the GIMP had no good docs, it would be a struggle. And that's the case with a lot of the serious Linux audio apps. I expect this to change too; but I don't know how quickly.

A second issue, IMHO, concerns the latency numbers I quoted earlier. To achieve them, I had to apply two patches to the kernel source and recompile. The average user is not going to want to recompile their kernel to get great performance. Perhaps the 2.6 kernel improvements no longer require this; OK. But the second thing is that in order to get these kind of numbers, JACK has to be run with realtime scheduling. This, in turn, requires the user to do one of two things: run JACK as root (in which case, all the apps that will interface with it, like Ardour or Hydrogen or Rosegarden, have to run as root too); or enable "POSIX-capabilities" inheritance in the kernel so regular-user apps can obtain realtime scheduling privileges. The former is scary -- it's always scary to run complex applications as root -- while the latter requires a kernel recompile (even with 2.6) and is a security risk.

The third issue is that while some application areas (HDRs, MIDI sequencers, effects plugins, mastering tools, etc.) are doing well, others (e.g. samplers) are in their early stages, while still others (e.g. notation editing) are almost completely lacking. The last point bears repeating: there are projects (like Rosegarden or Noteedit) that give you some notation-editing capability, and there are projects (like Lilypond) intended to produce excellent typeset scores; but there's nothing on Linux that really compares to Finale or Sibelius at this point, and there are no projects I know of even in their birth stages that are intended to fill that gap.

I'm using Linux for audio work; I like doing so, and (through reporting of problems) like contributing to its continuing improvement. But it has a ways to go to equal or surpass Windows in this regard. I think it will get there eventually, but it certainly isn't there yet.