Hdparm does not work when trying to optimize HD performance

[muadib25]

Hello!

I run FC9 on a P4, using both the pre-compiled 2.6.26 and a self-compiled 2.6.31 kernel. My hadrdrives are 2xWD Caviar 80Mb, both on ext3 . The first is alone on the first IDE bus, and the second is together with a DVD-RW on the second bus.
trying:

# /sbin/hdparm -t /dev/sda

Returns:

/dev/sda:
Timing buffered disk reads: 164 MB in 3.02 seconds = 54.30 MB/sec

Without any flags it returns:

/dev/sda:
IO_support = 0 (default 16-bit)
readonly = 0 (off)
readahead = 256 (on)
geometry = 9729/255/63, sectors = 156299375, start = 0

But trying to alter I/O support, for example, with:

/sbin/hdparm -c1 /dev/sda

I get this:

/dev/sda:
setting 32-bit IO_support flag to 3
HDIO_SET_32BIT failed: Invalid argument
IO_support = 0 (default 16-bit)

Googling a bit, I came accross this link,

http://ubuntuforums.org/showthread.php?t=415057&page=3

which is for Ubuntu troubleshooting, but I am not so sure if
it can be applied in Fedora.

Here's the solution that a user proposes:

Re: feisty hdparm /dev/sdXX was /dev/hdXX
My Acer notebook has had the sam issue. I found a solution.

The problem is with the new "ata_piix" module. The solution is to blacklist this module and make a new initramfs:

1. Create a file, "blacklist-local" in /etc/modprobe.d! (sudo touch /etc/modprobe.d/blacklist-local)
2. Edit the file (sudo gedit /etc/modprobe.d/blacklist-local)
3. Write the following line in it, then save it:
blacklist ata_piix
4. Edit /etc/initramfs-tools/modules (sudo gedit /etc/initramfs-tools/modules)
5. Add the following lines and save it:
piix
ide_generic
ide_cd
ide_disk
6. Execute the following command: sudo update-initramfs -u
7. Update all files has /dev/sda (or sdb...) in /etc and /boot. You need to change them to hda (hdb...). These should be only the fstab and menu.lst files. But if you use UUIDs you don't need to modify anything.
8. Restart your system.

Be sure all edited files has a final blank line (enter).

After these steps you got old good hda disks and you can use hdparm as well.

I hope I could help. Best regards:
Adam

Has anyone encountered and solved this??

[stevea]

hdparm was written before both the libata and ata_piix, and these new drivers do not correctly report the bus settings and they do not allow you to change the bus settings. The bus settings are automatically selected and you have no control or real insight without poking at the interface.

So even tho it claims you have a 16bit bus and no DMA etc - THIS IS NOT TRUE. IT IS A REPORTING ERROR.
When you try to set the bus width is fails b/c the driver automatically selected that and you have zero control.

There is no problem to solve here. 54MBPS is a completely respectable performance figure for contiguous disk reads from previous generations of stat/pata drives.

FWIW I get...
~33MBPS from a SATA Seagate ST3160023AS (7200.5) circa 2002
~55MBPS from a SATA Seagate ST3160827AS (7200.6) circa 2004
~73-75MBPS from SATA Seagate 7200.10s (circa 2007/2008).
A friend reports
~100MBPS from Seagate 7200.12 (early 2009)

So there is nothing wrong with your disk performance.

To remove the upper layer disk drivers can be accomplished as your note suggests, but you aren't having a performance problem so it is foolish. Also note that you will need to revise the udevd tables and also build a new initrd (with mkinitrd) after you have jumbled the drivers.
===

Also examine the sdparm package and the sdparm command.

[muadib25]

Thanks Steve, that was quick

Ah, so I guess I'll leave this as it is. I started customizing my system for low latency audio operations after I had noticed the reports about 16bit I/O Support...I was also trying to activate a flag when I was going to recompile the 2.6.31 realtime kernel called ' Generic PCI bus-master DMA support', but I no matter how hard I tried, I couldn't find it in the config setup...

Btw, isn't SDPARM for Scsi? I checked earlier today, it doesn't work with IDE Drives...