Re: 3Ware 9x00 controllers

Hi,


On Wed, Oct 27, 2004 at 11:34:27AM +0200, Remco Barendse wrote:
> On Wed, 27 Oct 2004, Pasi Pirhonen wrote:
> 
> >Just to comment this part. I would not hurry too much with this
> >updayte. My initial testing shows some 30% performance degration with
> >this new driver + firmware. The driver actually is pretty much same as
> >in 'trasferring linux scsi-layer request to the firmware and back', but
> >there is this new (shouldn't this be available begining of summer?) BBU
> >(Batter Backup Unit?) support for cache, so the ~100MB of your cache
> >won't 'stay on it's way to the platter' when one does loose suddently
> >power from the box :)
> 
> Oooooops! The card is already flashed but ok.
> 

As i said, it's reverseable (thank god) by just booting the previous
version of the driver. That just flashes the previous version of
firmware back in while loading.

> 
> >Good thing is that one is able to boot back to the previous version of
> >driver. The driver just flashes the old firmware in at boot.
> >
> >I am actually doing some numbers now with this old firmware to be
> >compared to new one. I didn't do much 'logging' with this x86-64 box
> >having 3w-9xxx and the ia32-box isn't available for testing anymore
> >(there is almoust 800GB data already which i won't play with :).
> 
> Great, if you have the results let's both raise an issue at 
> 3ware about it, hopefully they'll improve the driver.

It's more like 10% for general (sequential read/write and random
read/write).

Mostly the sequentila writing being slower a lot.

before it wrote like 90+MB/s for single thread. That slowing down to
like 70+MB/s for 8 paralle threads. 

Now it's in range of 70-MB/s for single and some 65+MB/s for 8 threads.

I really have to make some more testing still tho, so let this be 'one
more shot update' for this.



> 
> If the box boots after install I'll post some benchmarks of my array with 
> 4 raptors.
> 

that should not make the sequential writes any faster as the fabric
itself seems to be saturating at about 100MB/s (i think that is
supposed to be the write speed for 9500-seris). 

Raptors being 10k drives maybe affect for random access patterns tho.



-- 
Pasi Pirhonen - [email protected] - http://iki.fi/upi/