Andi Kleen wrote:
>
> >
> > >
> > > Another issue is the hook needed to tell the journal
> > > about memory pressure (this all has nothing to do with the VFS)
> >
> > Why shouldn't it be a VFS operation?
>
> I would have used a global hook (register_memory_pressure_handler())
> Hmm, ok a super block vector would do too, but the global hook would
> be probably useful for other things too (e.g. networking could use it for
> its skb caches). Block device interface has nothing to do with the VFS
> imho (you may be using a different definition of VFS than me tho)
>
> -Andi
I think that I proposed it both ways during the discussions of this 6 months or
more ago.:-)
I have always intuitively felt that fully separating caches is non-optimal. I'd
like to experiment with making pressure on caches proportional to the square of
the current cache size times some cache specific constant. Won't know until it
is tried if that is the right formula, but it would be neat to try it....
It wouldn't be unreasonable to start with it being a VFS operation, and end with
it being global for all caches. If the other cache code maintainers are willing
to accomodate I'd be happy to see it made global and happy to do the reiserfs
portion of that.
Hans
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Jun 07 2000 - 21:00:24 EST