>> If we get a journalling implementation that's something it can take care
>> of, whether it's PATA (flush the cache) or SCSI, where it can force
>> access
>> to the disc.
>
> Flushing the entire cache is unnecessary and will kill performance.

Unless things have changed recently, PATA gives you no other option.

> We only need to force the order for the metadata, for everything else
> we want to allow the disk to order the writes for the best performance.

It really depends on what guarantees you want to provide around data
security. E.g: is it OK to leak data from one users files into another,
do we care about data consistency after a crash, and so on.

Andrew