Re: piixide write speed very slow - solved

[Stephen Borrill <netbsd%precedence.co.uk@localhost>]

On Thu, 10 Dec 2009, Thor Lancelot Simon wrote:
> On Thu, Dec 10, 2009 at 04:34:41PM +0000, Stephen Borrill wrote:
> > Turns out that this was simply down to the write-cache being disabled
> > (OpenBSD explicitly forces the write-cache on). _Why_ the write-cache is
> > disabled is less clear.
>
> Careful.  Modern IDE or ATA drives alsmost always ship with the write cache
> on.  Some turn it off when an excessive bad block rate or other error is
> detected.  I wouldn't blithely turn it on if I found it off.

I support SMART might have some clues if that is what was happening. 
However, this is on two different machines with three different models 
and makes of drive. The cache is also off extremely early on.

> What I have observed, though, on one Intel-chipset system, for reasons I
> do not understand -- and in both piixide and ahcisata mode! -- is that
> the drive on one channel seems to turn its write cache off approximately
> every 2 minutes.  It doesn't matter which drive; put it on that particular
> SATA channel, and eventually it will turn off its write cache!  So turning
> write cache on explicitly at boot time did not actually help.
>
> Are you sure that's not what you're seeing?

I don't think so, my drivecache script runs before fsck (i.e. right as 
userland is entered) and the caches are off prior to that point. After an 
hour of RAID rebuilds they are still on.

> I don't think you should check in the "drivecache" script in its current
> form for the reason I give above.  Also, it should be able to manipulate
> the read and write caches settings separately.

I did discuss that with other developers. Certainly on these drives, 
read-cache isn't alterable. However, it would be worthwhile addition if 
the script itself was considered safe.

--
Stephen