Hi Ann,

Thank you for the nice and informative read!

You wrote

> [...] But if you let the operating system choose
> the order of page writes all bets are off. [...]

I still see a problem, that each higher layer
can only hold its promises, when the lower layers
hold their's.

So the OS must promise not to re-order writes (which
should be set by ForcedWrites), so that FB promises
on data security work.

But the disk controller must then promise not to get
in the way, or the OS cannot guarantee not to re-order.

In an article about soft updates for the UFS filesystem,
I found the dubious remark, that write caching had to be turned
off for the IDE drives tested, whereas due to a better
implementation this wasn't necessary for the SCSI drives.

Not sure whether this is still on the grounds of causality.

Regards,
Peter Jacobi