poor ATA disk speed with ICH2

Kevin Oberman oberman at es.net
Thu Sep 9 16:14:09 PDT 2004

> Date: Thu, 09 Sep 2004 23:24:05 +0200
> From: =?ISO-8859-1?Q?S=F8ren_Schmidt?= <sos at DeepCore.dk>
> Kevin Oberman wrote:
> >>>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=3D3Dad0 of=3D3Dad2), I can clear=
> ly he=3D
> >>
> >>ar
> >>
> >>>the sound of the actuator moving the heads constantly toward the end o=
> f=3D
> >>
> >>>the backup. I assume that they are being returned to track 0 on a
> >>>repeated basis.
> >>
> >>There should be no seeks on a lone write to the raw disk, if there is=3D=
> 20
> >>you have HW problems as the driver doesn't issue any seeks at all then.=
> >>
> >>What disks are this BTW ?
> >=20
> >=20
> > This was RELENG_5 on the 6th at 22:55 UTC. cvsup from my local mirror,
> > so it may have been up to about 70 minutes old.
> >=20
> > The "singing" only happens when running V5. I have no such sound when I=
> > do the same thing with V4.=20
> I have no idea, there is nothing in the ATA driver that does extra=20
> seeking, so the seek behavior must originate somewhere else, ie=20
> something else issues disk requests besides your dd. You could=20
> instrument the code and have it write out all LBA's it writes to on ad2=20
> to see what gets sent to the disk.
> > The disk is an Toshiba MK4019GAX. 40 GB ATA100 5400RPM. I get the same
> > performance, though, when using an IBM (now Hitachi) 40GB, ATA100
> > 5400RPM drive. What I am seeing looks a lot like what Bjoern was seeing=
> ,
> > so it's not unique to this system.
> Hmm, do you have write cache en/dis-abled ?
> Just for the fun of it I dd'ed to from my laptop disk through the fs=20
> though as I want to keep my data, but that shoudl only make performance=20
> worse. Its an ich4 but that uses the same code as the ich2.
> atapci0: <Intel ICH4 UDMA100 controller> port=20
> 0x1860-0x186f,0x376,0x170-0x177,0x3f6,0x1f0-0x1f7 at device 31.1 on pci0
> ad0: 76319MB <TOSHIBA MK8025GAS/KA023A> [155061/16/63] at ata0-master=20
> UDMA100
>   dd if=3D/dev/zero of=3Dfil bs=3D512  (24474468 bytes/sec)
>   dd if=3D/dev/zero of=3Dfil bs=3D1m   (25091961 bytes/sec)
> thats copying for ~30s as well (reading doesn't make sense here).
> Both are close to what that disk can do, so there is nothing in the
> driver thats prohibiting reaching the disks performance...

Weird. So what changes between V4 and V5 to cause this? (Actually, I am
unsure when it "slowed sown". I suspect it was with ATAng, but I just
don't remember any more and I don't have time right now to roll my
system back to pre-5.1 to see. Maybe this weekend.

I DO have disk cache enabled. I figured that the chance of a sudden power
failure happening on my laptop are very slim. (That was before the
preemption problem interacted with ACPI and killed the power to my

Clearly, if the heads are moving, it is not after every write. The times
would be MUCH worse than they are. Also, when I copy my full disk, the
throughput increases fairly dramatically toward the end of the copy.

I'll admit that I'm baffled at this point. I will do more testing,
probably with ktr, but it will have to wait until I have a bit more

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