I just finished reading the entire HowTo on LVM (http://www.howtoforge.com/linux_lvm
and it is GREAT.. I understand it WAY better than before, but I just wanted to clarify and question best practices regarding how LVMs should be expanded..
In the example, there were four small partitions, arranged into 2 mirror sets (md0= sdc1 and sdd1, md1=sde1 and sdf1), and a volume group comprised of the two md devices.
To expand the VG, one drive at a time was removed from the mirror set, replaced with a larger disk, and the partition 1 was re-mirrored, and a partition 2 was created, filling the rest of the disk, and being mirrored to the alternate drive's second partition.. So, the end result is that each drive has 2 partitions, and there are FOUR md mirror sets, all added into the VG.
I understand how this worked, and that no data loss occurs, but I'm curious if there's a performance disadvantage, or a scalability issue with leaving the old partitions, and adding new partitions, rather than creating a new md device with the new larger drive with one large partition, and then move the data off of the smaller drive onto the larger one, and end up with four drives, each with one partition? (kind of just a larger version of the initial setup)
To me, since we're talking about mirror sets, so drives can be replaced one at a time, the end state of only two mirror sets rather than four leaves a system which is at least easier to understand and troubleshoot, if not also having better performance just due to having less complexity, but perhaps under the hood, it makes no difference??
On a related question, are there any issues with using LVM with RAID5 devices over top?