Ripping CDs under FreeBSD and Linux 2.6

[savage]

Hey Guys,

It will be curious to see if anyone else has noticed this as well. I have FreeBSD on a hard drive in my current box. Under FreeBSD, it is not uncommon for me to rip music off of my IDE CD/DVD combo drive at about 20x (or grip says 20x).

When I boot back into windows, using IDE on 2.6.9-gentoo-r1, I rip audio at about 4 - 5x if I'm lucky. When I compiled with IDE-SCSI, my ripping speed went up to about 9-11x and seems pretty steady there.

So, the question is, is there anything I can do to improve the ripping off of CD speed under linux? Does anyone have any great ideas?

I've messed around with hdparm (and yes, turned on DMA). The ripping I am speaking of above is pulling the audio from the CD and throwing it into wav files. I am using the builtin cdparanoia in grip with all the functionality turned on. Any takers?

[geniux]

You could try ripping with abcde, it's cli though, but it's damn fast
_________________
AMD Athlon64 X2 4200+ AM2
MSI K9N SLI Platinum, Enermax Liberty 500W
1GB RAM Crucial DDR2 667MHz, MSI nVidia 7600GS 256MB
400GB + 250GB Samsung SATAII HDD
Gentoo - BeyondSources 2.6.19-20

[savage]

Hey folks,

Here's what I have tried:

compiling gentoo-development-sources with / without preempt;
compiling gentoo-development-sources with / without ide-scsi;

I've tried:
linux-2.6.10-rc1-mm1
linux-2.6.9-ck2
linux-2.6.9-gentoo-r1

I'm working on compiling linux-2.6.8-ac2 right now, and will post back if I have any luck with it.

I've double checked my numbers, and it turns out that the ide-scsi that I mentioned before wasn't that much of a speed up. Seems like under linux no matter what I do, I only rip at 3-4x speed; under FreeBSD it is 18-20x using the exact same tools. This is definately odd.

I recall back in my Debian days (and 2.2 - 2.4) I used the preempt with low-latency patches and that really gave linux the edge; perhaps I will recompile my gcc to not use ntpl and go try 2.4 with those patches again and see how it does.

I would really appreciate any ideas anyone has here - this is really odd and I would like to find the bottom of it.

Basically, I have several questions coming out of this.

a) Does the linux kernel have any problems with IDE devices that cause them to slow down so much?

b) If not all IDE devices, is it my chipset VT82cxxx?

c) Do these problems persist with other chipsets and on SCSI as well?

[savage]

Okay - no success here all the way around. Does anyone else that uses cdrecord have marginal performance under linux (i.e. 3-4x) instead of ~20x?

So, I've tried linux-2.6.9-ac2 now also, with no success. Also, I didn't mention it above, but I tried different schedulers, APIC on and off, etc. I've been compiling and rebooting for the better part of a full day now.

Anyone have any ideas?

[savage]

man - more stuff, but now I will put this down for a few minutes -> read until tomorrow

okay - have verified that the versions of cdparanoia were the same.

i.e. with cdparanoia -V

they are both III release 9.8 on March 23, 2001

used the time command under linux for song #1 on my test disc:

real time was 39.844

used the time command under FreeBSD for song #1 on my test disc:

real time was 11.15

Song length was 3:35.

i.e.:
Linux speed: 5.396x (wow! this is fast compared to what I normally get!)
FreeBSD speed: 19.28x

This was with custom compiled kernels under Gentoo, the GENERIC kernel under FreeBSD.

Also, no the disc is not scratched - it is practically brand new.

[sindre]

Maybe you should try posting this on the linux kernel mailing list.

[altorus]

This does ring a bell. I've noticed in the past that freebsd does have phenomenal ide cd read speeds for these sort of operations.

I'll look into it again when i have time, but just call it one of the good things about freebsd