> >From previous discussion I have impression that people think that with forced writes on
> inserting 10 rows into table (in one transaction) will generate 10+ i/o operations, which imho is not true.

I leave that to someone else to confirm or deny?

> When inserting large amount of data, they must be finally physically
> written to file anyway, so the main difference between forced writes on
> and off is that with forced writes off the OS is free to change order
> of physical writes, thus (potentially) minimizing disk rattling.

Perhaps in *nix, but in Windows it has been proven that the
PHYSICAL write to disk of large quantities of data does not
happen until windows actually gets round to it - by being
told.

> Thread for writing pages to disk when server is idle could be handy,
> but such thread would not be probably activated when importing large amout
> of data anyway.

The discussion here is how to get windows to ACTUALLY write
to disk rather than holding the information in it's own disk
cache until something else forces it to disk ( or not if the
power fails ) even after the information has been commited.
In my own case, we had instances of Windows crashing because
of a sound card fault, and all tickets issued, processed and
commited in the previous hours ( up to 4 ) where lost when
the machine was restarted. We now close the connection and
re-open it every 10 mins or so and only lose that window,
but having now replaced the problem sound cards, the system
is stable again anyway. A thread may be writing the pages,
but is windows taking any notice?

--
Lester Caine
-----------------------------
L.S.Caine Electronic Services