Re: Proposal: B_ARRIER (addresses wapbl performance?)

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Tue, Dec 09, 2008 at 10:14:25AM -0800, Jason Thorpe wrote:
> 
> On Dec 9, 2008, at 2:16 AM, Manuel Bouyer wrote:
> 
> >I'm not even sure FUA is necessery, if the ordered tag is working
> >properly (i.e. the device doesn't merge different writes in its cache,
> >which would break ordering).
> 
> Repeat: Ordered tags have nothing to do with whether or not a write is  
> committed to stable storage.

Agreed.

> But that is precisely what you want for  
> a write to a journal.  FUA is the mechanism that is specifically  
> intended to get the behavior you want for a journal.

But I don't get why we want to have the journal on stable storage *now*.
From what I understand, what we want is to have the journal entry for
this transaction to stable storage *before* the metadata changes start to hit
stable storage, and the journal cleanup hits stable storage *after*
all metadata changes are on stable storage. I can't see what FUA brings
us here, as long as writes to stable storage are properly ordered.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--