[lug] RH Linux download + hdparm

[Calvin Dodge]

On Wed, Oct 17, 2001 at 06:58:00PM +0000, Greg Horne wrote:
> So I noticed the command 'hdparm' in the previous e-mailand decided to play 
> around.  Boy is that fun!  Testing read speeds on different drives gave me 
> some interesting information.
> 
> Pentium 200 2.5 gig hard drive (RH 7.0)
> 'hdparm -t' = 4.83 MB/sec
> 
> K6 500 10 gig hard drive (RH 6.2)
> 'hdparm -t' = 3.83 to 4.6 MB/sec
> 
> Do the different distros on them make any differece?  Do the hard drive 
> brands play a big role in this?  Are there any safe tweaks to speed up 
> access times?

There are supposedly a number of tweaks (invoking various UDMA methods, changing the default read count, etc.), but the only one I've found to 1) make a difference and 2) be relatively safe (YMMV) is enabling DMA (the latest Red Hat kernels do this automatically).

Try "hdparm -d 1 /dev/hda", then see if that makes any difference in the transfer rate. I suspect you'll see a significant difference with your K6. I also suspect you'll see little difference with the P-200.

I don't know about different distros (I'm pretty much a Red Hat guy (not to be confused with the "red guy" from Steve Martin's "Writing Is Easy")), but in general it seems that the newer kernel versions understand more chipsets (like that difference between the 2.4.7 and 2.4.9 kernels).

The drives do make a difference - Mom's 500 Mhz K6-based system (with a Tyan S1598 motherboard) has a 5 gig Micropolis and a 30 gig IBM.  The Micropolis maxes out at about 4 megs/second, while the IBM (a 5400 RPM model) does about 16-17 megs/second.

The same model IBM in our old work server (a P-133 with a 430 TX chipset) transfers at about 8 megs a second - there I think it's obvious that the IDE controller is the bottleneck.

IIRC, with DMA off the IBM drive ran at about 4-5 megs/second (regardless of CPU speed or board age).

And - as mentioned before - the 5400 RPM Maxtor drives on our new work server show about 32-33 megs/second (yes, they're connected with those fancy new 80-wire IDE ribbon cables).

I hope that's enough data (without boring the people who already know all of this).

Calvin

-- 
Calvin Dodge
Certified Linux Bigot (tm)
http://www.caldodge.fpcc.net