Jason Thorpe wrote:

> On Jul 7, 2006, at 4:33 AM, Magnus Eriksson wrote:
> 
> >   Perhaps one of you wizards could tell me what it is that makes it  
> > absolutely true that "replaying the journal" == "writing to disk".   
> > I just don't see the connection.
> >
> >   Shouldn't it be possible to chug through as much as is needed to  
> > get the filesystem into a consistent state (or even the entire  
> > journal) and just cache the result?
> 
> Yes, that is precisely what I meant when I said "replay the journal  
> in-core".

I've only been half-following this journaling thread, so this may have
come up and I missed it...

Does that limit the journal size to something that fits in RAM?  Or do
we only keep enough of the replayed journal to allow us to mount then
do the "real" replay to disk that fsck or whatever does (which is maybe
another way of saying what Magnus asked?)?

Also, what is a "typical" journal size?

Simon.