LSI53C1030 (Fustion MPT) performance

LSI53C1030 (Fustion MPT) performance

[Karl Pitrich]

hi,

i got a new ibm intellistation z pro dual xeon (6221-49G) with on board 
Fusion MPT chipset (LSI53C1030) and fast U160 disks.

2.4.20-8 (redhat) and 2.6.0-test9-vanilla 
(each customer compiled minimal kernels) 
both yield very poor disk performance.

i didn't do specific benkchmarking, but mv'ing a 3GB $HOME to another
partition takes at least 4x the time as on my old P4 workstation with
IDE drive.

is poor performance with this controller a known problem?
could that have to do something with smp?

in 2.6.0, the driver's cvs-versions seem to match the ones in the
sources offered by LSI for download.
lkml-archives and google weren't all too helpful.

any info/help apreciated,


 / karl


PS: in the course of this i studied dmesg and found:

  calibrating APIC timer ...
  ..... CPU clock speed is 3066.0234 MHz.
  ..... host bus clock speed is 133.0314 MHz.

is 'host bus clock speed' != front side bus speed?
this shows up also on other p4 machines with much faster FSB than 133
mhz.

Re: LSI53C1030 (Fustion MPT) performance

[Andrew Morton]

Karl Pitrich <pit@0xff.at> wrote:
>
> hi,
> 
> i got a new ibm intellistation z pro dual xeon (6221-49G) with on board 
> Fusion MPT chipset (LSI53C1030) and fast U160 disks.
> 
> 2.4.20-8 (redhat) and 2.6.0-test9-vanilla 
> (each customer compiled minimal kernels) 
> both yield very poor disk performance.
> 
> i didn't do specific benkchmarking, but mv'ing a 3GB $HOME to another
> partition takes at least 4x the time as on my old P4 workstation with
> IDE drive.
> 
> is poor performance with this controller a known problem?
> could that have to do something with smp?
> 
> in 2.6.0, the driver's cvs-versions seem to match the ones in the
> sources offered by LSI for download.
> lkml-archives and google weren't all too helpful.
> 

The 2.6 driver runs nicely in single-disk mode but has been reported to run
like a dog in RAID mode.  I do not know if the 2.4 driver has the same problem.

Re: LSI53C1030 (Fustion MPT) performance

[Karl Pitrich]

On Mon, 2003-11-24 at 20:57, Andrew Morton wrote:
> Karl Pitrich <pit@0xff.at> wrote:
> >

<snip>

> > 
> 
> The 2.6 driver runs nicely in single-disk mode but has been reported to run
> like a dog in RAID mode.  I do not know if the 2.4 driver has the same problem.


well, i don't have a raid or mirror container configured.
the driver from 2.6.0-test9 (vanilla) runns nicely in fact, but is really, really slow,
my UDMA IDE machine is quite fast compared to the MPT.
(FYI: colleagues having the same machine running win2003 also tell me the disks are relatively slow)

further, i kindly heard from Dan Creswell <dan@dcrdev.demon.co.uk> that several 
people experience performance problems with this controller and that LSI is also 
aware of that and eager to fix this.


/ karl