[vox-tech] Debian bf2.4 base kernel (disable framebuffer, enable ide DMA)

Troy Arnold vox-tech@lists.lugod.org
Sun, 23 Feb 2003 17:21:40 -0800


On Fri, Feb 21, 2003 at 02:21:48PM -0500, msimons@moria.simons-clan.com wrote:
> Took a while to send this, since I wanted to double check what I wrote.
> On Tue, Feb 18, 2003 at 09:17:27PM -0800, Troy Arnold wrote:
> > On Tue, Feb 18, 2003 at 04:46:03PM -0500, msimons@moria.simons-clan.com wrote:
> > > - get rid of the VGA VESA frame buffer mode that is enabled in that kernel.
> > > 
> > >   This weekend I discovered that you can turn off the vesa frame buffer
> > > with a boot time command line option:
> > > ===
> > > video=vga16:off
> e > ===
> > 
> > Yup, it's documented in Debian's lilo.conf along with:
> > #vga=ask
> > 
> > Which gives you some other modes to choose from.  The mode that I
> > happen to prefer is not in the list:
> > vga=0x305
> 
>   I'm am a little puzzled...

Nope, i'm the puzzled one.  My fingers got ahead of my brain again:

> vga= and video= do completely different things.

I see that now...

> > There's also the 'hdparm' command:
> > hdparm -d 1 -m16 /dev/hda
> > -d 1 == turn on DMA
> > -m16 == turn on multi-sector IO
> > 
> > These two options vastly improve performance on modern IDE drives.
> 
>   Yes, I thought so too... I used to use: -d 1 -m 16 -c 1 -u 1 
> 
>   In my own testing I found that multi-block mode doesn't appear to 
> improve performance if dma mode is available.  If dma mode is available 
> enable that and it's as good as you get.  If dma is _not_ available 
> (because it's a really old IDE drive), then multi-mode helps some.

Makes sense, since (as I understand it) the whole point of DMA is to
bypass the CPU, while the point of multi-mode is to transfer more stuff
per interrupt.  I did some more testing and got results similiar to
yours: i.e. no real benefit if DMA is already enabled.  (I only tested
on two machines, using hdparm -t)

However, if I have an instance of vmware running, multimode helps
significantly:
 Two tests both with vmware running, but idle. (avg of 5 runs)

-m16
Timing buffered disk reads:   = ~ 28.32 MB/sec
-m0
Timing buffered disk reads:  = ~ 23.07 MB/sec

More important than numbers -- on a purely subjective level, -m16 helps
"a lot" with system interactivity if I have the guest OS copy a bunch of
mp3's.

-troy