Re: Ideas for TUX2

[Daniel Phillips <phillips@bonn-fries.net>]

On Wednesday 04 July 2001 09:53, Ph. Marek wrote:
> >> Well, my point was, that with several thousand inodes spread over the
> >> disk it won't always be possible to update the inode AND the fbb in one
> >> go. So I proposed the 2nd inode with generation counter!
> >
> >The cool thing is, it *is* possible, read how here:
> >
> >  http://nl.linux.org/~phillips/tux2/phase.tree.tutorial.html
>
> Well, ok. Your split the inode "files" too.
>
> Hmmm...
> That sound more complex than my version (at least now, until I've seen the
> implementation - maybe it's easier because it has less special cases than
> mine).

Yes, it's more complex, but not horribly so.  It's a lot more efficient, and 
that's the point.

> And of course the memory usage on the harddisk is much less with your
> version as you split your inode data and don't have it duplicated.

Yep.

> Well, I hope to see an implementation soon - I'd like to help, even if it's
> only testing.

See you on the list.  By the way, if you want to help out right now, could 
you run some benchmarks my latest early flush patch?  See "[RFC] Early Flush 
with Bandwidth Estimation".

--
Daniel