Leyne, Sean wrote:

> If you have a battery backed RAID with a reasonable amount cache
> allocated to writes, then there is NO REASON to not always run with
> Forced Writes = ON.
>
> The RAID card will completely isolate the 'cost' of Forced Writes.

So, the solution is to use "special" hardware that in effect caches the
forced writes in a safe manner? Where "safe" refers to protection
against power loss.

A thought: I assume the largest bottleneck is that FB has to write to s
few different areas of the file, e.g. list of active transaction, index
pages, etc.

Now, I know FB can use multiple DB files. Would it be possible to build
the engine to put one kind of object in each file, and then put each
file on a separate physical drive? This would, i presume, reduce head
seeks/thrashing. I know this is not possible today, but would it be a
viable way to improve performance with forced writes on? I can spot
several potential problems with this myself... it was just a thought.

Any other bright ideas on:

1. How to configure FB or the DB to reduce head seeks on commits with
forced writes on?

2. How to improve FB itself (future development) to do the same?

I note that IB and FB are the only DB:s I've seen that do so many head
seeks on commit. So how do other engines do it?

Kjell
--
--------------------------------------
Kjell Rilbe
DataDIA AB
E-post: kjell@...
Telefon: 08-761 06 55
Mobil: 0733-44 24 64