On Fri, Oct 24, 2014 at 12:09 AM, Steven Haigh <[log in to unmask]> wrote:
> On 24/10/2014 3:08 PM, Orion Poplawski wrote:
>> On 10/23/2014 10:02 PM, Steven Haigh wrote:
>>> For the record, I did manage to get this going:
>>> ## Wipe the disk completely and use the whole disk in ext4 config.
>>> zerombr
>>> clearpart --all --drives=xvda
>>> part / --fstype="ext4" --onpart=xvda --label=root
>>> bootloader --location=none
>>
>> Just an aside - no swap?
>
> Nope - If I need it, I do a file on disk instead... - or another
> partition which is also 'full disk' in the LVM...

I can recommend this approach. The "pre-allocate swap as part of your
disk image" used by anaconda  has led to some very, very inefficient
configurations for virtual machines. My one concern is that I'm not
sure you've succesfully aligned your partitions with 4096 byte block
boundaries, It makes a big performance difference if your back end
storage is NetApp or other setups that use bulky, modern 4096 byte
block disks. Do check the alignments on your disks with "fdisk -l"
after setup.

Mind you, I've previously written tools to pre-partition a disk, drop
a big old OS image tarball into the swap space, and used *that* to
uncompress an OS image onto the other partitions. It was very, very
fast indeed, and worked on roughly 13,000 hosts in one month, 12 years
ago, and was *much* faster than the old "replicate a disk image"
technology.