Re: [Dump-users] dump and snapshot

[Stelian P. <st...@po...>]

On Tue, Feb 18, 2003 at 04:38:15PM -0500, Andrew Liebeskind wrote:

> Well, since I seem to have taken on the role of EVMS-booster here, I might as well keep it up. :-)

that's the spirit ! :)

> EVMS does have a somewhat broader scope -- it provides a mechanism for managing standard disk partitions, LVM containers and volumes, md (software-RAID) objects, and a few more esoteric things like OS/2 volumes.  It also lets any of them be converted into EVMS-native objects, where you can take advantage of the EVMS features like bad block relocation or, most importantly in this case, snapshots.

I stand corrected.

[...]

> > No, it is not possible. You can snapshot only an existing LVM (logical)
> > volume.
> 
> And this is where I think EVMS makes it a bit easier if you're trying to add it to an existing install.  EVMS can handle your existing partitions just fine, as "compatibility volumes"; it can also convert them to EVMS-native volumes by shrinking the filesystem slightly (without loss of data, assuming it's a shrinkable filesystem type) and adding EVMS metadata to the end of the volume.

I didn't know that EVMS provides 'in place' conversion from existing
partitions to EVMS volumes. That's indeed a very good point in its
favour.
> 
> Once you've got EVMS native volumes, you can snapshot them by using any other object defined in EVMS as backing store; I use an empty disk partition with no filesystem for this purpose.  My backup scripts use the EVMS CLI to create a snapshot for each filesystem, use dump to backup the snapshot, then destroy the snapshot and create the next one.  

No different than LVM here...

> My experience with EVMS has been very good thus far, so I do recommend it as a general-purpose storage management solution for Linux, and particularly for the problem of snapshot backups you're dealing with here.

I agree. I should also add that EVMS is comercially backuped by IBM,
is an old and proved technology (EVMS was ported from AIX).

There is however a slight question mark concerning the future of
LVM/EVMS in the 2.5/2.6 kernel (none of two was accepted for 
inclusion, but a third, more generic solution, called 'Device Mapper'
was choosed, and LVM/EVMS should be poretd to work upon the DM layer).
However, I do believe that when the 2.6 kernel will be out (and 
stable for general use), the situation will be solved.

Stelian.
-- 
Stelian Pop <st...@po...>