Re: WG: Re: Bits (Nybbles?) from the Vancouver release team meeting

[<foo_bar_baz_boo-deb@yahoo.com>]

Thanks for working with me on this. Your help is appreciated. I've
combined the benchmarks, some system info, and some other hard disk /
controller / driver issues together below. I attempted to label the
sub-sections clearly to make this readable.

When I saw your first numbers, I too was surprised with the low scores.
However (luckily) I get better numbers. I did some experimentation with
the disk flags as you (very wisely) advised on one of the disks for
comparison purposes and it majorly increased the score.

Some things:
I used DMA66 thin ribbon / ground cables.
One of my drives is not "stock".

The dmesg outputs motivating my original (misinterpreted and/or poorly
phrased) IDE controller rant:
hdc: dma_timer_expiry: dma status == 0x60
hdc: DMA timeout retry
hdc: timeout waiting for DMA
hdc: status error: status=0x58 { DriveReady SeekComplete DataRequest }
hdc: drive not ready for command
hdc: status timeout: status=0xd0 { Busy }

Do you suppose I ought to be blaming Seagate or Sun's firmware for the
drive instead?; I just assumed it must be CMD's fault because it
usually is. Given you are the SPARC expert of note in the Debian arena,
your opinion on this issue is greatly valued. I am concerned this might
at some point frobnicate my boot drive from the snafu state to the
fubar state, so to speak. :-)

dmesg config strings:
hda: WDC WD400EB-00CPF0, ATA DISK drive
hdc: ST39140A, ATA DISK drive

benchmark log:
root@qux:~# /sbin/hdparm -c3 -m16 -d1 -X34 /dev/hda

/dev/hda:
 setting 32-bit IO_support flag to 3
 setting multcount to 16
 setting using_dma to 1 (on)
 setting xfermode to 34 (multiword DMA mode2)
 multcount    = 16 (on)
 IO_support   =  3 (32-bit w/sync)
 using_dma    =  1 (on)
root@qux:~#

root@qux:~# /sbin/hdparm -t /dev/hda

/dev/hda:
 Timing O_DIRECT disk reads:   46 MB in  3.10 seconds =  14.82 MB/sec
root@qux:~#

root@qux:~# /sbin/hdparm -t /dev/hdc

/dev/hdc:
 Timing O_DIRECT disk reads:   10 MB in  3.03 seconds =   3.30 MB/sec
root@qux:~# /sbin/hdparm -c3 -m16 -d1 -X34 /dev/hdc

/dev/hdc:
 setting 32-bit IO_support flag to 3
 setting multcount to 16
 setting using_dma to 1 (on)
 setting xfermode to 34 (multiword DMA mode2)
 multcount    = 16 (on)
 IO_support   =  3 (32-bit w/sync)
 using_dma    =  1 (on)
root@qux:~# /sbin/hdparm -t /dev/hdc

/dev/hdc:
 Timing O_DIRECT disk reads:   42 MB in  3.03 seconds =  13.87 MB/sec
root@qux:~#

--- "David S. Miller" <davem@davemloft.net> wrote:
> On Tue, 26 Apr 2005 18:35:37 -0700 (PDT)
> <foo_bar_baz_boo-deb@yahoo.com> wrote:
> 
> > What kind of benchmark did you run?
> > It would be sort of silly if I didn't do a similar test.
> 
> /sbin/hdparm -t /dev/hda
> 
> which does an uncached O_DIRECT 20MB read from the IDE
> disk, it's the real disk bandwidth not a cached number.
> 
> Also, try "/sbin/hdparm -c3 -m16 -d1 -X34 /dev/hda" if
> the performance stinks even worse than the 6.6MB I'm
> getting.  DMA tends to not get enabled by default unless
> the disk is in the IDE layer white list, the Seagate's
> that came standard in Ultra5 and Ultra10 systems just
> so happen to be in that white list.