Re: Proposal: B_ARRIER (addresses wapbl performance?)

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Mon, Dec 22, 2008 at 04:44:03PM -0800, Bill Stouder-Studenmund wrote:
> > I don't think it's so simple. I think there's also metadata writes that
> > needs to happen *before* journal writes.
> 
> The only place I see FUA on non-journal writes would matter would be to
> make it easier to roll the journal.

If you don't journal file data, but want to ensure that a newly
allocated block ends up on the disk before it is referenced, you need to
ensure the order.

> I think the advantage of FUA over write cache off is if we aren't 
> journaling file data. If we are journaling everything, then write-cache 
> off + simple tags will suffice.

Full journalling makes a number of things easier, as you can go with two
phase ordering. You don't have to disable write caching if you ensure
that it is flushed e.g. after writing half of the journal and throw away
the old journal data as written after that.

Joerg