At 11:11 AM 21-03-02 +0800, Kenneth Foo wrote:

>I imagined forced writes to simply issue an OS I/O flush to disk after
>each commits. If that's the case, then the load should be on the OS,
>not IB itself.
>
>I might be wrong, as I never looked as the FB sources before.
>As I said...I "imagined" forced writes to be that way :-)

Ah...imagination...

The "flush-to-disk" experience with Win is not so cosy. We've had
first-hand reports here that Win actually never flushes to disk until you
terminate ibserver; and, when it does, of course, it flushes every write
that has been done since ibserver was last started.

Forced Writes On is the only safe thing on NT. (Of course, the low-end Win
plats don't support deferred writes, anyway). Lost write cache is probably
the most common cause of FB/IB data corruption on Windows (and it is not
EASY to corrupt FB/IB data). It was one of Firebird's first priorities to
make FW "On" the default, for this reason.

cheers,
Helen

All for Open and Open for All
Firebird Open SQL Database · http://firebirdsql.org ·
http://users.tpg.com.au/helebor/
_______________________________________________________