Maya,

> Ann Wrote:
> >For what little it's worth, fifteen years ago, setting forced writes
> >on had a significant effect on performance on Solaris, AIX, HP/UX,
> >and DG/UX. That's why it's an option.
>
> .. and we have been telling all our users (all on FB 1.5 on Windows)
> they MUST switch forced writes on to help prevent data corruptions,

but

> some of them refuse to, swearing it is too slow to use if forced

writes

> are off!
>
> So A. a lot of checking he forced writes option after reports of data
> corruption, and B. a lot of arguments over speed for nothing ;-P

I think clarification is required.

Forced Writes on **Windows** has ALWAYS worked and is still STRONGLY
recommended.

Forced Writes on **Linux** has NEVER worked due to a known (but very
hard to find in the bug list) bug in the Linux Kernel.

v2.1 RC1 is the first build/release which has the necessary code changes
to enable the functionality. (I'm not sure if the change has been or
will be backported to earlier versions)

> Question: will FB 2.0 (where forced writes is now working) actually

have

> any effect on either speed or database corruption? Or is the effect
> negligible..?

By definition Forced Writes can slow down performance, since the OS will
ensure that writes are flushed.

The real performance impact can be mitigated by a good disk controller
with write-back cache -- to the point that it is not even noticeable.


Forced Writes, though, are REQUIRED in order to **absolutely protect**
against database corruption.

Without Forced Writes, corruptions can and will easily occur on Windows
in the event of a server shutdown/restart while the database is active,
and occasionally on Linux (the OS is better in this regard).


Sean