Fedora 15, new and exciting plans

Dennis Jacobfeuerborn dennisml at conversis.de
Sun Nov 14 02:23:31 UTC 2010

On 11/14/2010 12:41 AM, Richard W.M. Jones wrote:
> On Fri, Nov 12, 2010 at 06:26:48PM +0000, "Jóhann B. Guðmundsson" wrote:
>> *DE could consider switching the default to use EXT4 directly without
>> LVM. [1]
>> 1. http://fedoraproject.org/wiki/Features/NoDefaultLVM
> The "Detailed Description" seems contradictory:
> | LVM provides very little benefit for most Fedora users, at the cost of
> | performance and complexity:
> |
> | * Certain filesystem features (ext3 barriers) are unavailable when run
> |   on top of LVM.
> Isn't this just a bug which should be fixed?  (I actually thought this
> had been fixed already)
> | * Software RAID performance is greatly reduced when layered on LVM.
> But the stated task is to get rid of LVM except for "experts in
> storage administration" (from the next section of the same document).
> Who will presumably be the only ones wanting Software RAID.  The
> non-experts won't know anything about Software RAID, so they won't be
> affected by this performance problem with LVM.

Can someone point to specific details about this? I did some benchmarking a 
while ago of raid-1 vs. raid-5, raid-1 plain vs. raid-1 with lvm, etc. and 
LVM didn't really show up as a performance issue at all.

> | * LVM partitions are not automatically assembled by the desktop systems.
> I'm not sure what this one means.  "assembled" as in what happens when
> you spread a VG over multiple block devices?
> Anyway, I think LVM is jolly useful:

I've used plain partitions for a long time because lvm always looked weird 
to me but then I looked into it and nowadays I don't want to live without 
it. The ability to have the logical partitioning indepedent of the physical 
storage is a must-have for me.

> - You can expand the root filesystem (eg. into spare space or
> across block devices).
> - You can live pvmove filesystems from one device to another.

That one actually saved my ass once on a 48 disk 30TB storage system 
because the controller was acting up.

> It may be that the tooling is not there to make these features
> available for non-experts, but that's a problem with lack of tools,
> not with LVM.  Partition tables are horrible and inflexible in
> comparison to LVM.
> Can we at the very least have some numbers backing up the supposed
> performance problems?

Yeah, in my benchmarking I couldn't really confirm this so if there is a 
problem I'd like to see some specifics too.


More information about the devel mailing list