[parted-devel] [PATCH 6/6] libparted: fix optimal IO alignment

John Gilmore gnu at toad.com
Fri Dec 21 12:15:44 UTC 2012


> > I forgot one important reason for 1MiB, many people asked for
> > device independent partitions layout if possible. So you can dd(1)
> > your partition table and data from on one device to another device
> > without care about I/O limits.

I agree 100%.

> People really shouldn't be doing that.  It wastes time copying unused
> space and causes problems with duplicated UUIDs, as well was
> preserving any existing filesystem fragmentation, and of course, the
> size matching issues.  

There are very good reasons to copy entire partitions.  Such as making
a backup copy of a damaged partition before attempting to repair it.
Or copying an entire partition off of a disk that's giving SMART
warnings or otherwise showing imminent signs of failure.  Frequently
these very good reasons occur in the middle of a system administration
crisis -- at the wrong time for some self-appointed expert halfway
across the Internet to have built the software so that it won't do
what the sysadmin asks it to because "people really shouldn't be doing
that".

Copying entire devices is similar to copying partitions -- especially
for single-partition devices, which are very common -- and should be
equally well supported.

> Trying to do this with a disk with mdraid or
> lvm partitions on it will really confuse the hell out of mdadm/lvm.

Sounds like a bug in mdadm and lvm.  They should at least ignore
duplicate partitions, rather than getting confused.  But they are a
minor part of the market (most devices don't contain any mdraid or lvm
partitions), so we shouldn't let this bug control what we do for
devices in general.

	John



More information about the parted-devel mailing list