[BBLISA] Definition of dump levels?

[John Orthoefer]

There are several factors you need to consider when making a backup 
routine.  

The two extremes are always do a full, and never do a full.  

Always do a full, you have 1 backup which gets you to a know time 
(everything was done between 9pm and 1am) and the backups are a fixed 
size (1.25TB in use your backup is going to take 1.25TB+/-) and for long 
term static files, if it's bad on a tape you have additional copies of 
it.   The downside is the time transfer the data (bandwidth off the disk 
and bandwidth to the tape tend to be the gating factors), and the amount 
of tapes you need.

Incremental (after doing a full, assuming incremental meaning files 
since the last full/lower level) only, you win in that you use smaller 
numbers of tape, and bandwidth off the disk/to tape.  You lose in that 
only have one copy if there is a problem with the tape.  Also your 
incremental asymptotically approach the size of a full, so as the full 
gets further into the past your gains over a full diminish.

What people here are calling differentials just fall into Incremental 
category for me.   Because it's an incremental since the last 
incremental, now your recovery time goes though the roof and is also 
dependant on N tapes being good.

So depending on disk read bandwidth, tape write speed, processes which 
have to been shutdown (read databases), number of tapes in the cycle, 
and disaster recovery requirements, you get different  backup schemes. 

You have to think about this with your site.   Most of the time you have 
several different policies for different device classes.  ie, the netapp 
gets fulls every month, and incremental every week and differentials 
every day (lots of data, high speed reads, % of change is small from day 
to day, want to be able to do fast recoveries if you lose the device and 
for the most part you use snapshots for individual file recovery.)  
Desktops get full every 4-12 months, and daily incremental (small % of 
change day to day, don't care about recovery speed, small amount of  
"data")  database server, full every week (whole file changes, can only 
"stop" the database Saturday night from 1-3am)

johno

ps. yes there are ways around the database problem, the classic being 
the triple mirror (it's was only an example.)