FreeBSD on SSD

[RW]

On Sat, 28 Jul 2012 20:22:11 +0200 (CEST)
Wojciech Puchar wrote:

> > The read-cache idea is very sound, mainly because by using it this
> > way Seagate would not have to create a special set of instructions
> > for installing and using the HDD.
> 
> I don't think that this drive cache is smart enough to really cache
> needed things and not flush that cache with useless data too often.
> 

From what I've read the cache is divided into two part. One part holds
blocks that are consistently read during boot-ups. The other part holds
short runs of data that have been read off the disk with long access
times.

There's an oldish FAQ here: 

http://forums.seagate.com/t5/Momentus-XT-Momentus-Momentus/Momentus-XT-Frequently-Asked-Questions/m-p/52701

Note question 6

  "Internal tests were conducted using a script that mimics 250GB of
  data written to the drive every day for 5 years, and the Flash in
  Seagate’s unique solution showed no signs of degredation.  If the
  Flash ever fails, the data is never lost because it is mirrored to
  the disk and the drive would continue to operate as a traditional
  7200RPM drive with 32MB of DRAM cache."


> 
> > My final question would be :
> > 
> > Seeing as the HDD only has a SATA connector, this would mean that
> > the SSD part already has a memory control device that regulates
> > access to that sector, whether it is a plain read-cache or not.
> > This would imply that FreeBSD could communicate with the HDD
> > normally, through the SATA connector, just like any regular HDD.
> 
> indeed.
> 
> Such a drive is a good idea, but complete lack of documentation (how
> it operate) is not.
> 
> You have to guess how this SSD-cache works because it is not
> documented.

Well, it is intended to be completely transparent.

> 
> the other thing is erasing data. You want to sell that drive and
> clear your data by
> 
> dd if=/dev/zero of=/dev/disk bs=1m
> 
> but does it clear SSD cache?

That isn't guaranteed to clear any modern disk since they started using
spare sectors

> i don't think so. Someone sophisticated 
> enough would perform raw read of cache chips and get cached data,
> which can actually be the most important part 

If you're that paranoid use geli, or don't sell.