Dell SAS 5i/mpt driver[was: possible issues with Dell/Perc5 raid]

Matthew Jacob lydianconcepts at gmail.com
Wed May 16 17:07:29 UTC 2007


It wouldn't surprise me in the slightest that ext3 is better albeit
bursty for writes. That's the Linux aggressive write-behind and
ext2/ext3 for you in a nutshell. It's also quite likely true that in
many cases that reads and I/O scheduling is a lot better in Linux.

If you have time to run some benchmarks, using something like dt on a
FreeBSD raw partition with lots of small sequential writes and using
the (now deprecated) 'raw' bound devices in fc5 might get more of an
apple-apple.

On 5/16/07, Randy Schultz <schulra at earlham.edu> wrote:
> On Tue, 15 May 2007, Matthew Jacob spaketh thusly:
>
> -}filesystem? ext3?
>
> Correct.
>
> -}
> -}I want one of these failing machines *in my lab*.
> -}
> -}On 5/14/07, Randy Schultz <schulra at earlham.edu> wrote:
> -}> On Fri, 11 May 2007, Matthew Jacob spaketh thusly:
> -}>
> -}> -}Time frame to resolution involves getting a machine into my lab that
> -}> -}evidences the poor performance and the time to sort out what the
> -}> -}problem is and fix it. Remote access doesn't work for me on this one
> -}> -}(contact me offline as to why).
> -}>
> -}> Hey Matthew,
> -}>
> -}> FWIW, I installed fedora core 5 on the box.  The difference was significant.
> -}> Writes, though bursty, increased significantly.  Reads increased by 1-2
> -}> orders
> -}> of magnitude.  I've attached the output of blogbench for fc5 FWIW.
> -}>
> -}> Being new to all this, what would be the next step?  Should I file a bug
> -}> report?  Or perhaps I should be asking what your timeline is?  All I have to
> -}> offer for help is remote access, which in this case doesn't work for you, or
> -}> perhaps testing as my C skills are fairly pathetic.  ;>
>
> --
>  Randy    (schulra at earlham.edu)      765.983.1283         <*>
>
> Rain puts a hole in stone because of its constancy, not its force.
>   - H. Joseph Gerber
>
>


More information about the freebsd-performance mailing list