Re: RFC: ATA to CAM integration patch

Mike Tancsa wrote:
> At 05:20 PM 7/2/2009, Mike Tancsa wrote:
>> But, going back to the original i386 image, with the boot blocks
>> reinstalled and using your latest patch, it seems to work! (however,
>> the same 300sec delay due to the cdrom ?  )
> 
> At first, I thought something was a miss speed wise, but it looks like
> this hardware is either having issues, or something is wrong in general
> as its the same no matter which driver is used.  Usually the speeds are
> much quicker than whats below on block writes
> 
> 
> Seeker 1...Seeker 2...Seeker 3...start 'em...done...done...done...
>               -------Sequential Output-------- ---Sequential Input--
> --Random--
>               -Per Char- --Block--- -Rewrite-- -Per Char- --Block---
> --Seeks---
> Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU 
> /sec %CPU
> 1        4000 39798 20.8 39725  4.3 17776  3.4 40269 27.1 42085  4.2
> 255.8  0.6
> 2        4000 38827 20.3 40116  4.4 18068  3.4 40227 27.3 42266  4.3
> 244.8  0.5
> 3        4000 39748 20.8 40166  4.4 17952  3.3 40192 27.3 42259  4.3
> 243.4  0.5
> 4        4000 39855 20.8 40066  4.4 18017  3.3 40206 27.1 42401  4.2
> 264.2  0.6
> 
> 1=AHCI in bios, AHCI.ko loaded
> 2=AHCI in bios, plain old ata driver used post patch
> 3=IDE in bios, plain old ata driver used post patch
> 4=IDE in bios, plain old ata driver from the cvs

This test looks inadequate. there is almost no modern SATA HDDs having
only 40MB/s of linear read/write speed. Usual values now are 60-100MB/s
and they should be reached with almost any working driver. Can you try
simple `dd if=/dev/ada0 of=/dev/null bs=1m count=1000`?

To obtain any measurable benefit from NCQ usage you should have many
random requests to the drive running simultaneously. Not sure how this
specific test works. Also NCQ depends on effective disk firmware to
realize that benefit.

> Note, with 2 dmesg shows
> acd0: FAILURE - READ_BIG ILLEGAL REQUEST asc=0x64 ascq=0x00
> acd0: FAILURE - READ_BIG ILLEGAL REQUEST asc=0x64 ascq=0x00
> 
> the boot process then hangs for about 5 seconds, and then proceeds
> 
> with 3, the boot process hangs a total of about 2 min.

If you have issues with old driver also, then it is probably some drive
specifics, but not a bug of the new implementation. There was no changes
to the old ATA.

-- 
Alexander Motin