poor ATA disk speed with ICH2

Kevin Oberman oberman at es.net
Fri Sep 10 07:38:16 PDT 2004

> From: "Angelo Turetta" <aturetta at rainbownet.com>
> Date: Fri, 10 Sep 2004 11:38:03 +0200
> >----- Original Message ----- 
> >From: "Kevin Oberman" <oberman at es.net>
> >Sent: Thursday, September 09, 2004 10:28 PM
> >
> > While the test is running, the disk being written to "sings" with the
> > frequency somewhat dependent on the size of the write. On read, I get
> > silence. When I copy my full disk (if=ad0 of=ad2), I can clearly hear
> > the sound of the actuator moving the heads constantly toward the end of
> > the backup. I assume that they are being returned to track 0 on a
> > repeated basis.
> This sounds a lot like thermal recalibration seeks, to me. Except it's
> something modern hardware is supposed to need very rarely, if not at all...
> Might be the effect of some power-saving setting intruduced in 5.x ?
> (or maybe the new ATA code is actually so fast it's overheating your disks
> :-) :-)

Recals should take much longer than what I'm seeing. I have not dealt
with disk innards since back in SMD days, but back then a recal was a
very slow operation. These are fast. It is causing only about a 40% drop
in performance and only for writes, not reads.

I have an idea that, since the actuator activity is not with every
write but is very consistent, that it is happening only when the drive
switches tracks. Instead of just moving from one track to the next
(which is such a slight head movement that it is almost impossible to
hear), it is jumping back and forth between tracks. By track, I mean
physical cylinders on the drive. Because there is so much more data on
outer tracks, that would account for the decrease in the frequency of the
tone as the heads move out and the fact that it transitions from a tone
to pulses as the disk copy nears its conclusion. 

It may even be some oddity in dd(1) and not the system, at all. Of
course, this does nothing to explain what Bjoern saw with bonnie. That
is software that has little in common with dd(1).
R. Kevin Oberman, Network Engineer
Energy Sciences Network (ESnet)
Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab)
E-mail: oberman at es.net			Phone: +1 510 486-8634

More information about the freebsd-current mailing list