"William L. Jones" wrote:
> At 05:37 PM 9/4/00 +0100, Stephen C. Tweedie wrote:
> >Hi,
> >
> >On Fri, Sep 01, 2000 at 09:53:54AM -0500, Russell Cattelan wrote:
> >
> > > So I haven't gone and actually looked at code but given the only way to
> > get to
> > > 16TB is through a volume manager, the process or re-mapping requests
> > > will bring any individual device under the 2^40 limit but still allow
> > indexing at
> > > the ll_rw_block level up to 16TB.
> >
> >No. LVMs still get passed 512-byte indexed requests. A LVM exports a
> >block device, and that block device is just as much subject to the 2TB
> >limit as the physical block devices underneath it are. The fact that
> >the LVM device is a virtual device, not a physical one, does not make
> >a difference, sadly.
> >
> >Cheers,
> > Stephen
>
> My haed hurts. Your are right. Just when I figured out how make a large
> XFS file system.
I'll join you in the aspirin cocktail.
It's incredibly frustrating to find things that obviously should have been
implemented correctly in the first place.
I would say it shouldn't be much trouble fixing LVM to index up to 16TB.
I think time would be better spent working on the kiobuf I/O, since
the 2 TB isn't the only thing wrong with the current buffer head I/O
implementation.
--
Russell Cattelan
cattelan@xxxxxxxxxxx
|