kernel 2.6t9 SATA slower than 2.4.20

kernel 2.6t9 SATA slower than 2.4.20

[Shaun Savage]

I have just compiled and installed kernel 2.6t9 on my RH9 / Asus A7N8X 
Deluxe.  I find the disk access is slower using the 2.6 kernel than the 
2.4.20 kernel.

To get it to work for 2.4.20 kernel I have to use
# hdparm -d1 -X88 /dev/hde
then the buffered disk read goes from 1.5M to 55M

On the 2.6 kernel the buffered disk read is only 16M

What do I have to do to increase the disk speed for kernel 2.6t9?

Shaun Savage

Re: kernel 2.6t9 SATA slower than 2.4.20

[Ed Tomlinson]

On October 27, 2003 08:11 am, Shaun Savage wrote:
> I have just compiled and installed kernel 2.6t9 on my RH9 / Asus A7N8X
> Deluxe.  I find the disk access is slower using the 2.6 kernel than the
> 2.4.20 kernel.
>
> To get it to work for 2.4.20 kernel I have to use
> # hdparm -d1 -X88 /dev/hde
> then the buffered disk read goes from 1.5M to 55M
>
> On the 2.6 kernel the buffered disk read is only 16M
>
> What do I have to do to increase the disk speed for kernel 2.6t9?

What happens with you really test?  ie. using programs (bonnie, iozone etc)
to test disk speed?  Is it still slower?  I think that hdparm does not tell the complete
story.

Ed Tomlinson

Re: kernel 2.6t9 SATA slower than 2.4.20

[Nuno Silva]

Hi!

Shaun Savage wrote:
> I have just compiled and installed kernel 2.6t9 on my RH9 / Asus A7N8X 
> Deluxe.  I find the disk access is slower using the 2.6 kernel than the 
> 2.4.20 kernel.
> 
> To get it to work for 2.4.20 kernel I have to use
> # hdparm -d1 -X88 /dev/hde
> then the buffered disk read goes from 1.5M to 55M
> 
> On the 2.6 kernel the buffered disk read is only 16M
> 
> What do I have to do to increase the disk speed for kernel 2.6t9?
> 

I bet it's the "beat to death in lkml" issue with readahead.

What's the output of cat /proc/ide/hdX?

Regards,
Nuno Silva


> Shaun Savage
> 

Re: kernel 2.6t9 SATA slower than 2.4.20

[Jeff Garzik]

Shaun Savage wrote:
> I have just compiled and installed kernel 2.6t9 on my RH9 / Asus A7N8X 
> Deluxe.  I find the disk access is slower using the 2.6 kernel than the 
> 2.4.20 kernel.
> 
> To get it to work for 2.4.20 kernel I have to use
> # hdparm -d1 -X88 /dev/hde
> then the buffered disk read goes from 1.5M to 55M
> 
> On the 2.6 kernel the buffered disk read is only 16M
> 
> What do I have to do to increase the disk speed for kernel 2.6t9?


Are you using CONFIG_SCSI_SATA in 2.6?

	Jeff

Re: kernel 2.6t9 SATA slower than 2.4.20

[Arve Knudsen]

On Mon, 27 Oct 2003 08:53:01 -0500, Jeff Garzik <jgarzik@pobox.com> wrote:

> Shaun Savage wrote:
>> I have just compiled and installed kernel 2.6t9 on my RH9 / Asus A7N8X 
>> Deluxe.  I find the disk access is slower using the 2.6 kernel than the 
>> 2.4.20 kernel.
>>
>> To get it to work for 2.4.20 kernel I have to use
>> # hdparm -d1 -X88 /dev/hde
>> then the buffered disk read goes from 1.5M to 55M
>>
>> On the 2.6 kernel the buffered disk read is only 16M
>>
>> What do I have to do to increase the disk speed for kernel 2.6t9?
>
>
> Are you using CONFIG_SCSI_SATA in 2.6?

Should this make a difference speedwise with SiI 3112? I have one Maxtor 
120GB, and one Seagate 120GB each attached to the SiI controller (Mobo: 
Asus A7N8X Deluxe). The Seagate (my main drive) is as slow as ever (since 
DMA was turned on by default, ~13MB/s), although hdparm reports better 
numbers for the Maxtor (~33MB/s).

Regards

Arve Knudsen

Re: kernel 2.6t9 SATA slower than 2.4.20

[Shaun Savage]

 >Are you using CONFIG_SCSI_SATA in 2.6?

No, but I am trying now.
GREAT is works,
but the disk went from hda back to hde


 >    Jeff

===================================================================

 >I bet it's the "beat to death in lkml" issue with readahead.

 >What's the output of cat /proc/ide/hdX?

it hda   see attached file "proc_ide_hda"

now the disk is hde again, do you want another proc files file?

 >Regards,
 >Nuno Silva

--------------------------------------------------------------------

 >What happens with you really test?  ie. using programs (bonnie, iozone 
etc)
 >to test disk speed?  Is it still slower?  I think that hdparm does not 
 >tell the complete
 >story.

No I have not but I will.   But I assume that from 55M to 15M something 
is not configured correct.

 >Ed Tomlinson




---------------------------------------------------------------------
 > I have just compiled and installed kernel 2.6t9 on my RH9 / Asus 
A7N8X Deluxe.  I find the disk access is slower using the 2.6 kernel 
than the 2.4.20 kernel.
 >
 > To get it to work for 2.4.20 kernel I have to use
 > # hdparm -d1 -X88 /dev/hde
 > then the buffered disk read goes from 1.5M to 55M
 >
 > On the 2.6 kernel the buffered disk read is only 16M
 >
 > What do I have to do to increase the disk speed for kernel 2.6t9?
 >

I

Re: kernel 2.6t9 SATA slower than 2.4.20

[Jeff Garzik]

On Mon, Oct 27, 2003 at 07:56:31AM -0800, Shaun Savage wrote:
> 
> 
> >Are you using CONFIG_SCSI_SATA in 2.6?
> 
> No, but I am trying now.
> GREAT is works,
> but the disk went from hda back to hde

hmmm, with CONFIG_SCSI_SATA your SATA drives should show up as
 /dev/sda not /dev/hde ...

So, you're still using the drivers/ide driver, it appears.

Regardless, it's most important to use what works for you ;-)

	Jeff

Re: kernel 2.6t9 SATA slower than 2.4.20

[Arve Knudsen]

On Mon, 27 Oct 2003 11:59:17 -0500, Jeff Garzik <jgarzik@pobox.com> wrote:

> On Mon, Oct 27, 2003 at 07:56:31AM -0800, Shaun Savage wrote:
>>
>>
>> >Are you using CONFIG_SCSI_SATA in 2.6?
>>
>> No, but I am trying now.
>> GREAT is works,
>> but the disk went from hda back to hde
>
> hmmm, with CONFIG_SCSI_SATA your SATA drives should show up as
>  /dev/sda not /dev/hde ...
>
> So, you're still using the drivers/ide driver, it appears.
>
> Regardless, it's most important to use what works for you ;-)
>
Excuse me, there's probably something I'm missing, but how do I use the 
SCSI_SATA driver for SiI 3112? I see the source file for it in the kernel 
tree (test9), but no option for it in menuconfig (I've enabled SATA under 
SCSI). Enabling the SiI SATA driver under ATA/ATAPI... compiles in the old 
driver am I right?

Thanks

Arve Knudsen

Re: kernel 2.6t9 SATA slower than 2.4.20

[Bob Johnson]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

The Sata driver worked well on my siimage, but lost around 30%-40% performance
according to tiobench.

			Bob

On Monday 27 October 2003 10:56 am, Shaun Savage wrote:
>  >Are you using CONFIG_SCSI_SATA in 2.6?
>
> No, but I am trying now.
> GREAT is works,
> but the disk went from hda back to hde
>
>  >    Jeff
>
> ===================================================================
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)

iD8DBQE/nV8zxJgsCy9JAX0RAoMDAKCkhWzaYoOf4PT4xO+8deABSZuvwwCfSss/
9rIz1it4fWpQ3HVzFcK/qls=
=mOy2
-----END PGP SIGNATURE-----

Re: kernel 2.6t9 SATA slower than 2.4.20

[Jeff Garzik]

On Mon, Oct 27, 2003 at 06:36:37PM +0100, Arve Knudsen wrote:
> On Mon, 27 Oct 2003 11:59:17 -0500, Jeff Garzik <jgarzik@pobox.com> wrote:
> 
> >On Mon, Oct 27, 2003 at 07:56:31AM -0800, Shaun Savage wrote:
> >>
> >>
> >>>Are you using CONFIG_SCSI_SATA in 2.6?
> >>
> >>No, but I am trying now.
> >>GREAT is works,
> >>but the disk went from hda back to hde
> >
> >hmmm, with CONFIG_SCSI_SATA your SATA drives should show up as
> > /dev/sda not /dev/hde ...
> >
> >So, you're still using the drivers/ide driver, it appears.
> >
> >Regardless, it's most important to use what works for you ;-)
> >
> Excuse me, there's probably something I'm missing, but how do I use the 
> SCSI_SATA driver for SiI 3112? I see the source file for it in the kernel 
> tree (test9), but no option for it in menuconfig (I've enabled SATA under 
> SCSI). Enabling the SiI SATA driver under ATA/ATAPI... compiles in the old 
> driver am I right?

For SiI, it requires CONFIG_BROKEN, because, well, the libata SiI
is "very alpha quality" right now ;-)

	Jeff

Re: kernel 2.6t9 SATA slower than 2.4.20

[Jeff Garzik]

On Mon, Oct 27, 2003 at 01:08:46PM -0500, Bob Johnson wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> The Sata driver worked well on my siimage, but lost around 30%-40% performance
> according to tiobench.

This is to be expected -- I use a "sledgehammer fix" for the Maxtor and
Seagate errata -- each transfer is limited to 15 sectors.

When I get a chance to clean up the SiI driver, the performance will
indeed increase by a large amount.

For now, for Silicon Image, I recommend using the
drivers/ide/pci/siimage.c -- assuming it works for you, of course.

The libata Silicon Image driver is marked with CONFIG_BROKEN because it
is fairly easy to lock it up (I need ack some more interrupts).

	Jeff

Re: kernel 2.6t9 SATA slower than 2.4.20

[Arve Knudsen]

On Mon, 27 Oct 2003 13:15:29 -0500, Jeff Garzik <jgarzik@pobox.com> wrote:

> On Mon, Oct 27, 2003 at 06:36:37PM +0100, Arve Knudsen wrote:
>> On Mon, 27 Oct 2003 11:59:17 -0500, Jeff Garzik <jgarzik@pobox.com> 
>> wrote:
>>
>> >On Mon, Oct 27, 2003 at 07:56:31AM -0800, Shaun Savage wrote:
>> >>
>> >>
>> >>>Are you using CONFIG_SCSI_SATA in 2.6?
>> >>
>> >>No, but I am trying now.
>> >>GREAT is works,
>> >>but the disk went from hda back to hde
>> >
>> >hmmm, with CONFIG_SCSI_SATA your SATA drives should show up as
>> > /dev/sda not /dev/hde ...
>> >
>> >So, you're still using the drivers/ide driver, it appears.
>> >
>> >Regardless, it's most important to use what works for you ;-)
>> >
>> Excuse me, there's probably something I'm missing, but how do I use the
>> SCSI_SATA driver for SiI 3112? I see the source file for it in the 
>> kernel
>> tree (test9), but no option for it in menuconfig (I've enabled SATA 
>> under
>> SCSI). Enabling the SiI SATA driver under ATA/ATAPI... compiles in the 
>> old
>> driver am I right?
>
> For SiI, it requires CONFIG_BROKEN, because, well, the libata SiI
> is "very alpha quality" right now ;-)
>
Ok, thanks. So I might be better off with the older driver for now? From 
what I remember Andre Hedrick was promising a fix?

Regards

Arve Knudsen

Re: kernel 2.6t9 SATA slower than 2.4.20

[Jeff Garzik]

Arve Knudsen wrote:
> Ok, thanks. So I might be better off with the older driver for now? From 
> what I remember Andre Hedrick was promising a fix?


Yes.  Until CONFIG_BROKEN marker is removed, I recommend the older 
Silicon Image driver in drivers/ide.

	Jeff

Re: kernel 2.6t9 SATA slower than 2.4.20

[Shaun Savage]

Hi

I have removed the IDE SiI driver now the hdparm buffered disk read are 
now 15M again. So you are right the IDE does give better performance but 
the SCSI SATA does need to be enabled. I Think ;\

Now I am trying to iptables working, then SELinux

Thanks
Shaun


Jeff Garzik wrote:
> On Mon, Oct 27, 2003 at 01:08:46PM -0500, Bob Johnson wrote:
> 
>>-----BEGIN PGP SIGNED MESSAGE-----
>>Hash: SHA1
>>
>>The Sata driver worked well on my siimage, but lost around 30%-40% performance
>>according to tiobench.
> 
> 
> This is to be expected -- I use a "sledgehammer fix" for the Maxtor and
> Seagate errata -- each transfer is limited to 15 sectors.
> 
> When I get a chance to clean up the SiI driver, the performance will
> indeed increase by a large amount.
> 
> For now, for Silicon Image, I recommend using the
> drivers/ide/pci/siimage.c -- assuming it works for you, of course.
> 
> The libata Silicon Image driver is marked with CONFIG_BROKEN because it
> is fairly easy to lock it up (I need ack some more interrupts).
> 
> 	Jeff
> 
> 
> 
>