Re: Thin device provisioning

[Matthew Wilcox <matthew@wil.cx>]

On Sun, Aug 10, 2008 at 01:50:38AM +0100, Jamie Lokier wrote:
> Matthew Wilcox wrote:
> > I've spoken with a few Linux filesystem people.  They find it
> > significantly easier to send a single LBA/length pair at a time.
> > Modern filesystems try quite hard to keep fragmentation to a minimum, so
> > they don't expect a performance hit from sending multiple commands.
> > They're non-blocking writes, and the IO elevators can take care of
> > sending more important reads first.
> 
> Perhaps there are occasions when it's more efficient for the disk to
> process several LBA/length pairs in a single operation?
> 
> I.e. you send the first pair, the disk starts working, then you send
> the second, and that's not as efficient as doing both at the same
> time, which might translate to a single commit on SSD.
> 
> The general solution to that would be a 'CORK' operation, though,
> similar to TCP_CORK: this operation will be followed by others, you
> may start it now, but don't rush...

If you read what Fred Knight (of the T10 committee) wrote, he said that
they felt it would be easier for filesystems and drivers to use multiple
extents.  He didn't say anything about drives finding it more efficient.

-- 
Intel are signing my paycheques ... these opinions are still mine
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."