On Mon, 2 Feb 2009, Elizabeth Schwartz wrote:

> The obvious thing would be to go to full every 59 days and a
> 1,2,3,4,5,6,7 rotation, but I'm thinking we should get a little
> creative. If we do a full on Sunday and someone dumps 10G on Monday,
> that 10G will get backed up eight times.  I'm thinking the first week

I use an approach similar to the one you currently use.  Your proposed 
alternative is interesting.  I have a feeling I did originally consider it 
too but maybe I didn't look at it closely enough.

Another alternative is to move from a 7 to a 9 day incremental rotation: 
1->9.  Who said backup levels need to sync with the day of the week?

For comparison I'll touch on what I have done when setting up backup/DR 
for a number of organisations.  I didn't need to save money on backup 
bandwidth but made that point that DR is very rare compared to file 
recovery and backup, so aiming to maximise DR speed is not efficient. I 
concluded it is better to prioritise the design for a faster backup with 
rapid file recovery and disaster recovery being less important (in that 
order).

Speeding up DR may reduce down time from a disaster by 30 or 40% but the 
outage is still likely to be measured in hours or days.  If an 
organisation needs to recover from a disaster quickly then there are much 
better options than making the DR from tape faster.

Anyway, my approach is similar to your current approach:

I normally do a full once per month.  The full backup date is obtained in 
a pseudo-random manner[2].  If a full backup is not due then a level 2 
occurs on Monday, 3 on Tuesday ... 8 on Sunday. I deliberately avoided 
levels 1 & 9 so that I could "undercut" the incrmentals if I wanted to by 
running a level 1.  Level 9 can be run as required (eg, just before a 
major system update).

I'd have to review my notes but I concluded that there was no need to sync 
the incremental levels to the dates of the full backup for a given system 
- syncing them to the day of the week produced sabstantically similar 
results.

[2] I use custom scripts and generate the md5sum of the hostname and take 
a mod(days_in_month) of the first 5 characters of so.  This produces a 
pseudo-random distribution of the full backups across the month.  Even 
hosts that are named similarly will not necessarily have backups dates 
near one another.

Cheers,

Rob

-- 
I tried to change the world but they had a no-return policy
_______________________________________________
Tech mailing list
[email protected]
http://lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
 http://lopsa.org/

Reply via email to