trying to decide whether to use LVM or not

[Eric S. Johansson]

Craig Hagerman wrote:
> If you use LVM to span a couple disks (make a more than one 'appear'
> as a single hard drive) then if you have a failure or other problem
> with one disk you will likely lose everything - even the stuff on the
> good disc. THAT was the killer for me.

That's a really good gotcha.  It's interesting how the different types of 
failure tug us in different directions.  For example, the reconstruction time 
for a raid array implies that you want to have multiple small array so that if 
you lose one array, it's faster to rebuild.  The same principle applies to 
filesystem checks.  You want lots of small partitions to keep your fsck time 
down and narrow the region of damage should something fail.  But with small 
partitions and raid arrays, you have inefficient storage allocation as a 
partition approaches full, one must either shift data to another partition or 
enlarge the partition.

Logical volume management was designed to fix this problem to unify lots of 
small partitions and enable more efficient storage allocation.  But as you point 
out, lose one partition, you lose the whole thing.  Without any further counter 
evidence, I'm convinced that LVM isn't worth it even in a corporate environment 
(unless you're on a Sun or IBM platform).  So I'll have one partition, 250 MB 
large, raid1 with a spare drive.  I'll take a hit on rebuild time for raid1 and 
an fsck.  If I need any small partitions, there's always a loopback device.

Thanks Craig.  I appreciate the help.

---eric