[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: HOWTO - Speed up IDE HD's



hi ya

fun stuff

On Sat, 7 Feb 2004, Marc Wilson wrote:

> On Sun, Feb 08, 2004 at 12:07:26AM +0000, Svens wrote:
> 
> <hdparm suicide deleted>
> 
> Gee, let's turn another cluebie loose with a shotgun, and see if he can
> zorch his data.

yupperz... its fairly easy to lose your disk if you play with the 
wrong options on hdparm
 
> Let's not find out what modes his drives actually support, and let's make
> sure that the kernel can't reset it if it's wrong.  Let's unmask
> interrupts, without finding out if his hardware can deal with it.  Let's
> try to force DMA on, without finding out whether or not his kernel supports
> his IDE chipset.
>
> Never mind that the kernel can do 90% of this on its own, and a wonderful
> first step should *always* be finding out why it HASN'T.

i assume that assumption was made ... that one would check before applying
various hdparm options and checking against hw

most distro set the disk options to a  "safe option" to work with
most any mb/hd combo

Nano> To be honest, I tried what he said and got very little speedup.
Nano> I did try "hdparm -c" and saw that by default I was doing -c0.

yup.. if dma was already on .. doing those settings might not help
and definitely can hurt your disks if it didnt support -Xxx whatevr
you changed it to

if dma is on .. and your disks supports -c3 -u1 ... you can get some
10% - 20% speed improvement doing the "same tasks over and over"
	- say an infinite kernel compile for 24 hrs...
	a xp-1700 will be able to do the same number of passes as 
	a xp1800 

other ways to ide speed improvements
	- one ide disk per ide cable
	- do NOT mix different ata disks if you use 2 disks per cable
		atx33/ata66/ata100/ata133 should all be different cables
	- blah blah..

> People who write hdparm HOWTOs need to be hung up by their toenails.  I've
> yet to read one that tried to be an even minimally responsible resource.

fun stuff

c ya
alvin



Reply to: