Hi Alain,
I don't reallyknow, IIRC there is a timeout or idle detection for
flushing. It did not impact speed as a do-not-flush would do
It greatly improved databade system stability
As far as I know, both FreeDOS kernel and caches for it never
buffer multiple writes, waiting to flush them back to disk,
so you probably meant something else?
It COULD improve performance to bundle writes in some way,
such as the write-combining buffers of NWCACHE and possibly
SMARTDRV, so I have written from time to time about how even
a small amount of delayed writes could help performance. And
how that would have to be combined with well-chosen triggers
to flush before users could crash, reboot or power off, for
example "always flush when a program exits and/or the shell
returns to the prompt" or "always when DOS is idle and waits
for user input" etc.
However, there are no free delayed-write implementations for
FreeDOS yet and I think I did not even manage to have users of
NWCACHE collect spectacular data from experiments which would
simulate the suggested amount of delayed writing. If I just
have forgotten those, please point to the right place in the
mailing list archive so I can read the stats, thank you :-)
Regards, Eric
_______________________________________________
Freedos-devel mailing list
Freedos-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freedos-devel
