PCI-X SATA Card + Server Recommendation

[Jeremy Chadwick]

On Tue, Oct 28, 2008 at 10:24:30AM -0700, note at note2email.com wrote:
> > A large number of problems people report to the FreeBSD lists involve
> > Silicon Image controllers.  There are confirmed problems within certain
> > models of their SATA controllers which cause silent data corruption and
> > other issues, affecting Linux, FreeBSD, and Windows.  See "Product
> > Alerts" below, then try Googling "silicon image corruption".  I'm not
> > talking out of my ass.
> >
> > http://en.wikipedia.org/wiki/Silicon_Image
> 
> Which points to
> 
>         http://osdir.com/ml/ide/2005-03/msg00126.html
> 
> which says:
> 
>         "It's basically because of faulty SATA implementation of the
>         affected seagate hard drives combined with standard-compliant
>         but peculiar behavior of silicon image controllers."
> 
> So you blame Silicon Image for Seagate's bug.  Nice.

Wow, thanks, I see you really did your research before posting this.
Let's ignore the hundreds of posts between users on forums talking about
data corruption on SiI controllers with NON-SEAGATE disks, as well as an
equally large number of posts on Linux mailing lists with the same
facts.  I will not let you hold me responsible for Wikipedia's reference
material being sparse.


I'll also point you to similar 'misleading' information about nForce
controllers and Maxtor disks.  There are known incompatibilities with
some versions of nVidia MCPs and Maxtor disks.  The problem has to do
with NCQ support, but absolutely no one is certain if the problem is
with nVidia's chip or Maxtor's firmware.  nVidia has done absolutely
nothing about the problem, while Maxtor has documented the problem
and offers -- assuming you ask for it -- a disk firmware that works
around the problem by changing the NCQ implementation.  But the same
disks work fine on Intel, VIA, SiS, Promise, and even SiI controllers.
So who's to blame?  :-)  I really, *really* hope you get my point.

Let's not turn this thread a "usual BSD thread", where a bunch of
administrators sit around and do burn-outs in parking lots, getting
absolutely nothing accomplished other than gnashing of teeth.

> I have been using the 3512 with Seagate drives and NetBSD for
> several years with zero data corruption.  If FreeBSD has problems
> with Silicon Image controllers it isn't Silicon Image's fault.

I've provided enough evidence that the problem has NOTHING to do with
FreeBSD.  The problem is OS-independent.  Since you're explicitly
ignoring this fact, I have to classify this as trolling.

> Word is that the 3124 and 3132 are much better and faster than the
> 1st generation controllers such as my 3512.  They are documented,
> datasheets are available on the web, unlike some SATA controllers.
> I would consider them.
> 
> > This has been discussed recently on -hardware.  I will stand firm on my
> > statement: don't disable write caching on disks.
> >
> > http://lists.freebsd.org/pipermail/freebsd-hardware/2008-October/005450.html
> 
> Write caching on disks enabled:             data loss, fast writes
> Write caching on disks disabled, no NCQ: no data loss, slow writes
> Write caching on disks disabled,    NCQ: no data loss, fast writes
> 
> Sorry, but data loss is simply not acceptable.  When can we expect NCQ
> support for FreeBSD?

Matt Dillon and some others have explained why disabling write caching
is simply unreasonable, and why it gives people a false sense of
security ("if I disable write caching, my data will ALWAYS be written to
the disk!" is simply untrue).  It doesn't even matter if you have a
controller that has a BBU.

> The ability to turn the disk's write cache on and off is essential.

See kern/127717.  Despite not being an advocate of disabling write
caching, I've no problem extending tools/drivers to provide features.

> I haven't found a USB-to-SATA bridge that allows this, limiting
> their usefulness to testing new disks, mounting read-only or
> for data which is expendable.
> 
> > Besides, it shouldn't matter to you if the card
> > has RAID capability, because nothing forces you to use it.
> 
> The problem is that cards with real RAID are far more expensive.

The OP has specific requirements for SATA controllers.  The specifics
greatly limit what choices he has.  The fact of the matter is, there are
only two companies which make ""affordable"" consumer SATA controllers:
Promise and HighPoint.  I'm specifically excluding SiI from the list for
what at this point should be obvious.

-- 
| Jeremy Chadwick                                jdc at parodius.com |
| Parodius Networking                       http://www.parodius.com/ |
| UNIX Systems Administrator                  Mountain View, CA, USA |
| Making life hard for others since 1977.              PGP: 4BD6C0CB |