On Mon, Nov 23, 2020 at 10:32:18PM -0500, Jon LaBadie wrote:
> I did reply to the original message, but looking back it was addressed
> to you rather than the list.  In case it was overlooked, here were my
> regarding space:

I saw it, just absent-minded about replying sometimes...  Good that you
repeated it for the list, in any case, I figure.

> Did not know what VDO was, so I read a Red Hat description.  It seems
> to consist of 3 components each I question the value of for amanda backup.
> Hopefully someone with VDO experience can share it.

I first heard about VDO from a friend who (surprise, surprise) works for
Red Hat.  He also runs a hosting company on the side and has been very
happy with what VDO has done for his backup server, but his backup
system is pretty rudimentary (sounds like it's a homebrew rsync-based
solution), so I suppose his results may not be entirely indicative of
what to expect with amanda+VDO.

> Only one copy of duplicate blocks:  Were your files being backed up
> individually, as I do in a separate backup of my Home directory using
> rsync, this could provide a worthwhile savings.  But you will likely
> be merging your files into a tarball or a dumpfile.  The original
> disk block alignment will be lost and likely not even match in one
> day's tarball to the next.

Ah, yes.  I hadn't considered the tarball aspect of it.  I figured it
would be able to make good use of the deduplication if I have amanda
write "uncompressed" and use only VDO's compression, but you've got a
good point about merging the files still messing things up in that
scenario.

> LZ4 compression on the fly:  I don't know the cpu load for the server
> compressing 8TB of data daily.

I assume that the CPU load would be comparable (not identical, of
course, but in the same ballpark) for on-the-fly VDO compression vs.
amanda compressing the data at that level.

> There are points where amanda calculates how much space is left on
> the device based on it configuration-specified size and how much it
> has already sent.  Of course there is actually more space available
> because the compression occurs after amanda's involvement.  The
> difference may cause amanda to make less than optimal decisions.
> 
> Amanda administrators who use tape drive compression face the same
> problem.  I believe most over specify the size of the storage medium
> to allow more complete tape utilization.

Hrm, yes...  Now that you mention it, I do have foggy memories of seeing
this discussed when I was using amanda previously.

But, yeah, if block deduplication isn't going to be a significant
benfit (and it sounds like it probably won't), then it probably would be
better to skip VDO and have amanda handle the compression itself.

> In the distant past I backed up windows systems by mounting the drives
> on UNIX host.  Most often used NFS.

Any particular reason to prefer NFS over SMB?

> Windows, at least then, does not like a file to be opened by multiple
> processes.  So each backup included several files that did not backup
> because the file was already opened by another Windows process.  And a
> few system files were never backed up.

I'm pretty certain this is still the case.  One of the most annoying
misfeatures of Windows, IMO.

> Regarding backing up a KVM snapshot, would that mean that to recover
> one file you would have to take a new snapshot, restore the entire system
> from the backed up snap, copy the file to somewhere else, restore the
> new snap, copy the file to final location?

Pretty much, yeah.  That's why I haven't even considered snapshot-based
backups of my linux virts.  But, if it makes the Windows admin happy...

-- 
Dave Sherohman

Reply via email to