Alan McKinnon wrote:
On Sunday 07 October 2007, Florian Philipp wrote:
- You loose bit performance, but not much, you won't feel it without
benchmarks.
I very much doubt this. LVM is one extra layer between the filesystem
and the physical disk and it basically consists of a mapping between
the extents in the VG and exactly where they are on the volume. This is
nothing more than an elementary lookup table; on a 500G VG using 32M
extents this consists of precisely 15,625 entries, it can all be stored
in RAM and can consist of one pointer plus precisely one calculation to
determine the offset from the start of the table where the desired
extent lies.
Considering that RAM runs at many orders of magnitude faster than the
disk you are trying to get the data off of, the extra fraction of a %
overhead is not even worth trying to measure, let alone benchmark it.
Moving the heads just once more because of file fragmentation probably
takes longer than the entire LVM lookup
A year or two ago it was possible to measure a performance hit - for
instance we had some Supermicro PCI-X based machines with 3Ware RAID
cards where we could get (quoting from memory here as it was a while
ago) uncached sequential scan rates of about 1Gb/s without LVM and
somewhere in the region of 800Mb/s with it.
However that was then, and this is now - LVM and hardware have no doubt
improved , so it would be interesting to do the test again (if we do
I'll let you know).
I would think that for the OP's use case (i.e 1 disk on a desktop box)
there will be no measurable difference at all.
Cheers
Mark
--
[EMAIL PROTECTED] mailing list
