[PVFS2-developers] metadata record pre-allocation
rross at mcs.anl.gov
Tue Jul 27 13:46:18 EDT 2004
On Tue, 27 Jul 2004 neillm at mcs.anl.gov wrote:
> On Mon, Jul 26, 2004 at 07:35:49PM -0500, Rob Ross wrote:
> > Do you think it's necessary to export a function to deal with
> > preallocation? To some extent, trove knows as well as anyone whether or
> > not the storage system is busy; perhaps it would be better to simply have
> > trove kick off preallocation operations when its queue is empty and it is
> > below some low-water mark? It already has a thread most of the time
> > anyway, right :)?
> While other backends will probably have pre-allocation of some sort,
> I'm not sure what their interface would look like. Using the
> StorageHint space should allow full flexibility without having to work
> around anything. I think this should stay dbpf specific too because
> trove isn't threaded -- just the dbpf implementation of.
My bad; I wasn't being specific enough. I agree that this should be in
the dbpf implementation.
More information about the PVFS2-developers