[OpenAFS] Re: Backup hierarchies ?

Matthew A. Bacchi mbacchi@btv.ibm.com
Fri, 15 Feb 2002 09:15:49 -0500


>This one I would find irritating. Why should I define new hierarchies,
>if I decide to buy more tapes and keep them longer? Does not convince
>me.

The main reason to define different hierarchies is you will then have
distinct tape labels, which will make it easier to locate the one you
want.  Otherwise, you will have 8 sets of tapes all labelled the same,
and you would have to read the tape to determine when it was created.
One of our dumplevel trees looks kind of like this:
/sat1a/sun1a/mon1a/tue1a...
/sat2a/sun2a/mon2a...
/sat3a/sun3a/mon3a...
/sat4a/sun4a/mon4a...
etc.

Which will make the first tape done on saturday of week one be labelled
as sat1a.1.  You can have a 1 week long heirarchy if you want.

>> or by creating 1
>> eight week dumplevel.  
>
>How that? I want a full dump at the beginning of each
>week. Incremental for the other days.

This works like I mentioned yesterday, you would have only one level
0(Full) dump per 8 week period, and have weekly incrementals(level 1)
dumps based off that, then daily incrementals(level 2) based off
those. This allows you to have fewer full dumps of the whole cell.  In
a large cell where much data is static this can be beneficial.

>> Or, you can can create 1 one week dumplevel,
>> which will not expire.  
>
>It is mandatory for these dumps not to expire? I would like them to
>expire after 7 weeks....

I guess you have to understand that when a dump(tape) expires, it is not
automatically removed from the backup database.  You still have to do
a deletedump operation to get rid of the entry from the db.
Expiration is really only a method of allowing you to reuse the tape,
but in reality you probably don't want to do this without removing the
old tape label and data(we do this by writing EOF at the beginning of
the tape).  So, it is not mandatory that they expire.  And you still
have manual processing anyway whether they expire or not.

>Otherwise asked: Can I have several dumps (for several weeks) of the
>same dump hierarchy in my backup database? 

Yes.

>4) I am missing something that relates the actual incarnations of the
>full/incremental dumps ("objects") to each other. I suspect, this data
>structure is held in the backup database somehow, but how exactly?*)

I guess I really don't understand your question.  Maybe the problem is
that this system is highly configurable and can work in many ways.  I
think you need to experiment with it for a little while.  Design it
your way, do some backup for a couple weeks, and see how it works.
You will soon understand the relationship once you do some queries in
the backup db.  Start with the 'dumpinfo' command, to see a complete
listing of dumps(labels). Then do a 'dumpinfo -id' to view an
individual dump (both full and incremental) an see the relationship.
An important piece is the parent dump, and the level of the dump.  It
will become very obvious how all the levels relate at that point.

Hope this helps.
-Matt