I'd like to answer a couple of these things.
> > * load balancing across multiple scsi-channels (not implemented yet)
This is almost always at the driver level on the host. Since LVM is a sort
of Driver, much like MD/RD, then it would be as though MD/RD had that kind
of ability. Also, think of this like EMC's PowerPath(r). It will allow
overloading a single scsi channel, and overflow goes to a secondary channel
to ballance the bandwidth.
> > * Dynamic partitioning
This is something which can be done on the hardware side, but only at the
disk level. Partitioning is still an OS level function, LVM/EVMS allows for
this and other things too. Hardwar can do this if it is meant to, but that
is expensive. LVM/EVMS is not.
> > * Volumes spanning multiple physical devices
Think of something like an EMC frame, where you've got 200 PVs. What do you
do with them? You could MD, or RD, but LVM is much easier and closer to a
Veritas type volume management. Nothing new here.
> > * snapshotting.
It may feel like hardware, but ist still software, even if a dedicated box
is snapshotting for you. Moving the intelligence out of some storage box
allows you to go cheaper on that storage if the host is performing that
function instead. You do incurr increased costs however when you find that
you don't have enough bandwidth to snapshot + do your business at the same
time. More HBAs relieves this bottleneck.
> But how about performance?
> Aren't these things supposed to be handled by the hardware?
> It's the same thing as for soft vs hard RAID. Hard should be faster
> (but, yeah, soft is easier to play with).
> Florin Andrei
Systems Architect, CCNA