On 07/20/2009 02:40 PM, Don Quixote de la Mancha wrote:
A kernel developer, I think from Red Hat, said on a list that lvm is
2% slower than a physical partition.
Considering the obsession some folks have with performance, that seems
like an awful lot to give up for some flexibility which really may not
be at all helpful to some users.
The reason for the overhead is basically that when you send the
command over the wire to the actual disk, you have to give it an
absolute Logical Block Address - relative to the beginning of the
whole hard drive. Hard disk drives don't know from partitions or
logical volumes.
To convert a partition offset into a disk offset, you just add the
starting sector of the disk. To get that starting sector, you have to
look it up in a data structure that's maintained by the disk driver.
I don't know how LVM is implemented, but I imagine there are some
extra layers of indirection that enable that flexibility. The data
structures involved will be more complex, as will be the code.
They will also be more likely to be buggy as well.
I've been setting up a bunch of partitions to run virtual machines on,
for cross-platform development. While it's a PITA to keep
repartitioning my RAID 5, I figure the extra effort is worth it for
that consistently 2% faster disk I/O.
Don Quixote
quixote@xxxxxxxxxxxxxxxx
http://www.dulcineatech.com/
Dulcinea Technologies: Software of Elegance and Beauty
Thanks Don,
I was going to make those exact points, however for performance reasons
I go one step further and use the entire array without any partitions -
this gives perfect raid chunk alignment :), and was infact the only way
I could get the parallel and horizontal performance I needed (wanted).
Albert.
--
fedora-list mailing list
fedora-list@xxxxxxxxxx
To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list
Guidelines: http://fedoraproject.org/wiki/Communicate/MailingListGuidelines
