SOLVED Re: Suggestions wanted for

[Joe Emenaker <joe@emenaker.com>]

Joe Emenaker wrote:
> Ward Vandewege wrote:
> > So for what it's worth, the Sil3114 is *not* trash. Sure, it's not the
> > greatest sata controller in the world (the 3124 is considerably 
> > better, for
> > instance), but it does it's job.
> >   
> To add yet another data point, take a look at this thread in the 
> kernel devel list where they're discussing how to work around the 
> sil3114's quirks.
>
> http://www.ussg.iu.edu/hypermail/linux/kernel/0406.2/1821.html
>
> I'd like to draw special attention to the part where he says " Silicon 
> Image 311x is fully SATA compliant -- but it's the only controller 
> that sends odd-sized packets to the SATA device. That causes no end of 
> problems...".
So, I got my Promise SATA 300 TX4 in the mail on Friday. I installed it 
and all of the problems vanished. I was able to run my torture-test 
script (which constantly creates, copies, and deletes 200MB files...) on 
the RAID5 array (using 3 of the SATA ports) and a solitary drive on the 
4th SATA port all at the same time. I left them running for a few hours, 
with no data corruption.

Previously, with the Silicon Image 3114 card, I'd get corruption within 
about 1 minute.

To recap the steps I took:

- Torture-tested a PATA drive while torturing the SATA's, and the SATA's 
got corrupted and the PATA was fine.
- Changed from a 230W power supply to a 380W Antec, and the SATA problem 
persisted.
(At this point, I was fairly convinced that it wasn't a power-supply issue)
- Tried an identical SYBA Sil3114-based card, and the problem persisted
- Tried an Addonics Sil3114-based card, and the problem persisted.
- Tried a Promise SATA card (the first one I tried that wasn't based 
upon a Sil3114) and the problem vanished.

My personal conclusion is that there is either something wrong with the 
3114's or with the Linux driver for it (since I've not *noticed* any 
corruption problems with a Windows machine using a Sil3114). Whatever 
the reason, I'm going to steer clear of Sil3114-based cards from now on. 
The rest of you are free to do as you wish, of course.

- Joe

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature