Re: Multi-level incremental back-ups (was: Decremental backup: hard-disk space consideration)

Denis Corbin <[email protected]>
Newsgroups gmane.comp.sysutils.backup.dar.support
Message-ID <[email protected]>

On 15/02/2021 16:30, Stefan Müller wrote:
> Am 03.02.21 um 19:07 schrieb Denis Corbin:
[...]
>>
>> Thus, retention is thus very flexible in the strategy you expect to use.
> 
> Thanks for taking the time for your elaborate answer! My question had
> been almost as thoroughly worked out -- up to the title, that is. Sorry
> for any confusion this may have created. I had originally been thinking
> about using decremental backuups but their incompatibility with delta
> signatures made all related questions become irrelevant. Nevertheless,
> my main concern is hard-disk space, indeed.
> 
> The pointer to dar_manager is very much appreciated as I have not
> studied it yet but definitively will.
> 
> If I understand your description correctly, then the level-1 back-ups
> would contain multiple identical copies of files because they only
> depend on the level-0 back-up but not upon each other.

no, a given backup may it be full or differential does usually only
contain a single copy of a given file. The level-1 backup is a
differential one, it contains only data of file that change since the
level-0 used as reference was done.

> A file that was
> created immediately after the level-0-back-up would be contained in each
> one of them, wouldn't it?

It would be contained in level-0 backup (because it is a full backup),
as it has changed right after this backup it will be stored in the
level-1 backup too, either as a full copy (eventually compressed) or as
a binary delta based on the version stored in the level-0 backup

If it does not change any further, level-2 backup will just contain its
metadata (permission, date, ownership...)

But if it changes after a level-1 backup was made, it will be stored a
second time in the first level-2 backup that follows the change as well
as the following ones,

the next-month, the level-1 backup will also have a copy of it but
unless it changes again during that month, the level-2 backup will only
hold its metadata

> 
> Maybe I got the terminology wrong. I thought of incremental back-ups to
> be moving one level up at each stage as opposed to differential ones
> that always depend on a full back-up.

Well, you are right, the term differential/incremental are used
differently depending on people. One can define differential when it is
based on a full backup and incremental else. I usually do that too.

But here, when we consider the level-2 backups of a given month, if they
are all based on the level-1 backup made the first day of that month,
one could say they are "differential" backups, by opposition to the
situation where each level-2 backup takes as reference the backup of the
day before, in which case each level-2 can be seen as an incremental.

In the first case you restore level-0, level-1 and the level-2 of the
day you want.

In the second, you restore level-0, level-1, and all the level-2 from
the 2nd of the month to the day you want, in that order.

> 
> Using your terminology, the monthly back-ups should be level-0 for the
> original one, level-1 for the first incremental one, level-2 for the
> second etc. Within one year there would be up to level-12, within five
> years up to level-60.

difficult to understand which reference you take for each backup, seen
the numbering you use, I assume the backup of each month uses the backup
of the previous month as reference (except the first month which is a
full backup).

> 
> The daily back-ups would be like branches to these monthly ones. Thus
> there would be level-0.1 up to level-0.31 for the first month, each
> incremental back-ups, depending on each other. For the second month
> there would be level-1.1 up to level-1.31 incremental back-ups and so on.


OK I see,

> 
> The first monthly level-1 incremental back-up would basically wrap the
> preceding 31 daily ones (level-0.1  up level-0.31) into one single
> incremental back-up. This way, a single file change would be included
> twice into the pool, once in a daily and once in a monthly incremental
> back-up.

correct,

> 
> Since I envision almost "perpetual" or at least very long-term retention
> of the back-ups, I could delete the daily back-ups after a while and
> retain only the monthly ones.

Yes this is possible,

> Thereby, file changes would not balloon
> the pool as much anymore. After some epoch, the pool can be set aside
> and a new one can be started with the same back-up strategy.


That makes sense, yes


> 
> It must be obvious that I do not have any hands-on experience so I am
> very thankful for any comments!

dar_manager will be of greate help here to retrieve a deleted files
easily even long after it has been removed (by accident), so there is no
drawback here, human operation is a single command, though the
rebuilding of a file my imply a lot of backup if binary delta has been
used and the file changed a lot over time (thus has a binary delta in
each backup)

The issue is more annoying when it would come to restore the whole
system, in that case you would not use dar_manager, but would be more
efficient restoring the backups in order from level-0 to level-1-N then
from level-1-N to level-2-M (where N is the month and M the day of the
month you want ot restore the system state to).

This is a choice between the space requirement for backup and the time
to restore the backup, nothing is perfect. Added to this the probability
of a system global crash and you should take the best decision that
matches your context :^)

> 
> Cheers!
> Stefan Müller
> 
> 

Cheers,
Denis

_______________________________________________
Dar-support mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dar-support
OpenPGP_signature (application/pgp-signature, 840 B) - not displayed
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.