[Fedora-xen] Best practices questions

[Aaron Metzger]

Daniel P. Berrange wrote:

> 
>> LVM vs. regular partitions is the usual deal: LVM gives you more
>> flexibility.  You might find that you appreciate the flexibility even
>> more when you run guests.
> 
> LVM & partitions give best performance characteristics.
> 
> Regards,
> Dan.

Dan:

Thanks for your insight.

If, for example, a guest has separate /root, /boot, and swap do you 
recommend as best practice a separate logical volume for each or do you 
give just one logical volume to the guest for it to dice up into the 
various types of file systems?  Is there a significant performance 
difference between these two solutions?

For us newbies using virt-install or virt-manager, it prompts for at 
most one partitition (which can be a logical volume) so without editing 
of the config after the fact it naturally leads to the case of using 
just one logical volume for the entire guest with it dicing it up into 
/root, /boot, and swap or whatever.

So, for the case of just one logical volume which has been "formatted" 
during the guest install, this led to the question of how do I 
generically backup and restore that one logical volume.

Every piece of documentation that I find for creating an LVM snapshot 
and then creating a backup requires that the snapshot volume be mounted 
to perform the backup.  This led me on an adventure with device mapper 
and kpartx etc during which I was able to eventually get to and backup 
the inner guts of my guest but it seemed like there had to be a better way.

Is there a vanilla generic backup and restore procedure for a logical 
volume (a snapshot volume of the running guest logical volume) which 
does not require case-by-case knowledge of what type of file system(s) 
have been mapped to that logical volume?  If so, is there then also a 
way to do that backup in a manner which only stores the small amount of 
space actually used by the guest instead of the entire space allocated 
to the logical volume?  In other words, something that preserves the 
idea of sparseness?  This is probably a generic LVM question but I could 
not find my answer in the LVM docs either.


My lack of knowledge related to the above questions is why I am 
currently using image files but would love to switch to individual 
logical volumes for the performance benefits.

--
Thanks,
Aaron