On Sat, Oct 12, 2002 at 04:25:45PM +0200, Markus Plail wrote: > Hi Jörg! > > * Joerg Schilling writes: > >Note that the encoding speed is content dependant. Also note that due > >to a kernel design bug, Linux does not enable DMA for ATAPI drives if > >the sector size is != 2048 (as with RAW sector size 2448). This results > >in a high system load and may cause a buffer underrun even before the > >estimated maximum speed of the edc library has been reached. > This is not true, at least not as a general statement. As I reported in > a newsgroup, you are reading, I was able to burn a data CD with -raw96r > at 20x speed with my ATAPI CD writer with less than 5% CPU usage with > your closed source version. AUDIO CDs can be burnt with the same > options with less than 3% CPU usage even with the open source > version. So it's rather a bad setup than a general Linux problem.
How do you explain then that I for example can't write audio CDs any faster than 14x in Linux and the system is hopelessly bogged down due to this light IO task already at 8x writing? And in FreeBSD and Windows this is not a problem? On the other hand, this slowdown happens only if the writer is connected to VIA 686b. If it's connected to a CMD 649, Promise pdc20265 or Intel PIIX3 there's no problem. As enabling DMA for ATAPI devices leads to a Linux crash with the CMD and Promise chipsets, it's clear that PIO can go fast with a low system load, which is what you're seeing. Add to this that Alan Cox himself has stated that Linux doesn't use DMA for CD writing, I'd have to say you're wrong. I think Joerg put it very well in calling it a kernel design bug, as that's exactly what it is. The only good thing about it is that it doesn't affect many IDE controllers, the VIA 686b is the only one I know of. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
