inode checkpoints

Josh Boyer jdub at us.ibm.com
Tue Oct 5 13:20:38 EDT 2004


On Tue, 2004-10-05 at 11:45, David Woodhouse wrote:
> On Tue, 2004-10-05 at 18:07 +0400, Artem B. Bityuckiy wrote:
> > Hello David, Josh. What did we decided about the GC thread?
> > 
> > I offered to create checkpoints from the GC thread. Are there some 
> > reasons not to do so? There was one - the GC thread isn't always 
> > enabled. But we may require it be enabled if checkpoints are enabled...
> 
> Checkpoints are an optimisation -- so is the GC thread. I'm perfectly
> happy with the idea that killing the GC thread also stops checkpoints
> from happening.

Yep, me too.

> 
> > May be we create ICP during GC process, or on iget() call, or on
> > iput() call? But I have arguments against these...
> 
> I was thinking of doing it during a normal write to the file, or during
> garbage collection of a node belonging to the file.
> 
> But I think we can play with ideas on _when_ to call this once you've
> actually implemented the functions which does it :)

Right.  Lots of places it could be done.  Start with one and work from
there :).

josh





More information about the linux-mtd mailing list