[ale] boot parms ide=???

[Marvin Dickens]

On Thursday 20 February 2003 07:37 am, Danny Cox wrote:
> Marvin,
>
> On Wed, 2003-02-19 at 06:16, Marvin Dickens wrote:
> > hdparm -c1 -m16 -d1 /dev/hda
> >
> > This enables 32bit access and multisector reads. These items may already
> > be present in a compiled kernel. It depends on distro.
> 	Just one little nit to pick.  The man pages don't say this, only code
> inspection will reveal this.
>
> 	If you're using DMA, then -c{01} doesn't matter.  That only affects PIO
> transfers (16 bit words or 32 bit words), so by all means, if you're
> stuck with PIO, use -c1!
>
> 	The same is true for -mN.  The code has different paths depending on
> whether it's transferring single or multiple sectors.  However, when DMA
> is in use, that's all bypassed, and the multisector stuff is handled by
> the scatter/gather DMA code.
>
> 	All that to say that -c1 and -m16 don't hurt, but they don't do
> anything when -d1 is in force.
>
> 	It's also a good idea to run on -u1 IF YOU CAN.  Most modern chipsets
> work just fine with unmasking ints.  The effect of not setting it can
> cause seemingly unrelated things to happen.  For example, if you're
> burning a CD with -u0, and you have an established modem connection, it
> can drop (too many chars arriving, but ints are blocked, resulting in a
> dropped carrier).


Everything I know about hdparm is the result of trying to squeeze out every 
last ounce of performance from under provisioned Linux machines that are 
installed at customer sites. It works like this: The customer builds a Linux 
box out of an antique, performance is aweful and they call me in to tweak it. 
Another words, I read the O'rielly article and the man page.....  Personally, 
I run SCSI drives which seem to me to be faster and more reliable. But, with 
that said, it's kinda hard to ignore the fact that you can get a really BIG 
ide drive for very few dollars.


Best

Marvin
_______________________________________________
Ale mailing list
Ale at ale.org
http://www.ale.org/mailman/listinfo/ale