On Mon, Dec 10, 2007 at 05:56:09PM -0800, Tracy R Reed wrote:
James G. Sack (jim) wrote:
Perhaps I should have used virtualization instead of indirection, and
it's all handwaving, of course, but I was thinking that removing a layer
could improve performance, mainly latency, I suppose. And not less
important, I was thinking that there might be simplification leading to
better maintainability and fewer places for bugs to hide.
So far I have never seen a measure of latency induced by LVM or heard any
concerns that it is a performance bottleneck. Given that a few pointer
redirects in the block layer are FAR faster than the latency imposed by the
disk itself not too many people are concerned with LVM performance yet.
The only real complaint I have so far about LVM is that it doesn't seem to
yet pass write-barriers through to the underlying device. XFS notices this
right away, and disables the barrier. I haven't had any crashes to lose
data, and they fixed the XFS problem with the files full of nulls. But, it
is at least theoretically less robust on an LVM than directly.
But, at least for me, the LVM adds so much value to the system that it is
worth the risk.
Dave
--
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
