[linux-lvm] LVM 2
joe at fib011235813.fsnet.co.uk
Thu May 30 09:23:01 UTC 2002
On Wed, May 29, 2002 at 08:07:58PM +0200, Anders Widman wrote:
> > On Wed, May 29, 2002 at 02:16:44PM +0200, Anders Widman wrote:
> What exactly shown in those diagrams? :)
It's trying to answer the question 'how does io to a volume slow down
if I take a snapshot'. The *overhead* is plotted on the vertical scale,
in terms of the time taken to do the operation with no snapshot.
So if I use a chunk size of 16k we see that LVM2 has an overhead of
0.23, ie. the 'dbench 2' operation on a freshly snapshotted origin
will take 1.23 times as long compared to the snapshot not being
present. EVMS on the other hand takes 7.84 times as long.
If you find the numbers confusing just look at the raw results in the
first table. If you use 8k chunk sizes with EVMS expect 'dbench 2' to
take ~227 seconds compared to LVM2's ~15.
> > LVM2 by default uses the same metadata format as LVM1, so there really
> > isn't any conversion procedure. So try LVM2 for a bit, if you don't
> > like it (unlikely) revert to LVM1. Please note that metadata backups
> > and archives are held in a very different form.
> So I can just install LVM2 and continue as usual, and switch back if I
> get into trouble?
> And, what if I get into trouble? I can not backup
> all data. There is not enough space for that.
What happens if you get in trouble with LVM1 ? If that data is
important you should really back it up, all drives fail eventually.
> >> Also, is it possible to convert (with and LVM version) a spanded
> >> volume to a striped volume?
> > Not until we get the new pvmove infrastructure.
> So, when newer pvmove tools develops it would be possible to convert
> a spanned volume to a sriped (RAID0 or 5?) volume without loosing data?
RAID5 is probably 6 months away, but I would hope to be able to remap
a linear volume to a striped volume in the next month or so.
More information about the linux-lvm