Re: diff between incremental and archive backups
- From: Mladen Gogala <gogala.mladen@xxxxxxxxx>
- To: oracle-l@xxxxxxxxxxxxx
- Date: Sat, 14 Nov 2015 01:21:01 -0500
On 11/13/2015 01:59 PM, Zelli, Brian wrote:
So we were having a discussion about rman incremental backups. And
the question came up that if I do an rman full once a week and then
rman back up of the archive logs the rest of the week, that’s all I
need to do a point in time restore. I don’t have to do incremental
backups. Is this an accurate assumption?
Brian
Brian, the first thing to ask is what is the purpose of the incremental
backups. Hopefully, you are aware of the fact that backup is an
expensive operation. In order to not burden their users unnecessarily
during the peak working hours, which were usually during the working
week, people have resorted to incremental backups. The price to pay for
a quick backup during the week was longer restore and recovery time. If
your database croaks on Thursday, you need to restore the full backup
taken on Saturday and all the incremental backups from Sunday to
Wednesday. After that, you need to do a recovery.
However, today is backup the last line of defence. The foremost
mechanism to ensure quick recovery is duplication. Typically, production
database will be protected by some duplicating mechanism. There are
different mechanisms: storage and VM snapshots, disk replication like
SRDF, HUR (Hitachi Universal Replicator), SnapVault/SnapMirror, standby
databases or incremental backup in the FRA, which is constantly
recovered. Most of those duplication methods allow quick and painless
backup. If you backup your standby database, you can restore it to
production. RMAN catalog will recognize the standby and production as
the same database, because they have the same DBID. It is also possible
to mount disk snapshots or disk replica and do either a file system
backup or rman backup. Be aware that there is usually some scripting
involved or a product which can do that by itself. In case of a replica,
the consideration of slowing down users is eliminated. Users are not
using standby database. So, if you do a full backup of a replica every
day, you are not obstructing the business process, you are speeding the
recovery. The only obstacle left is storage. If you have 20TB database,
which is not particularly gigantic these days, than keeping 7 full
backups on disk would require 140TB of disk storage, which still costs a
pretty penny. That is where modern deduplication algorithms come into
play. They are typically much more efficient than compression and
achieve around 90% savings, as opposed to compression which achieves
around 70%. That efficiency comes at a cost, usually requires SSD and
10GB network, but you will need at least 10GB network any way if you
want to backup 20TB database in less than a week. So, with 90%
deduplication efficiency, your full backup shrinks down to 2TB, which is
about the same space as consumed by the incremental backup.
So, you have now solved the question of strategy: some kind of
duplication will be needed and a proxy backup of the replica, with
deduplication added into the mix. The only remaining question is now the
one of SLA: what is the time that your CIO will give you to restore a
20TB database. This database is likely powering a web application which
should be available 24x7x365. Twelve hours offline is usually
unacceptable. If PayPal of Ebay ever went offline for the full 12
hours, they would be destroyed by the competitors. So, switch to standby
or revert from snapshot will speed things immensely, but you will still
have to restore 20TB in 4 or 5 hours, which is quite a feat. For that,
you will need storage attached with a FC adapter and a good SAN, with a
lot of cache. Cheapo NAS storage cannot do the trick.
Long story short, you will need to spend a lot of money on hardware, SAN
licenses, Oracle licenses, infrastructure and alike. Forget the
incremental backups, almost nobody does that any more. If somebody is
doing incremental backups, that is because of ignorance, not because of
some "secret sauce" which will speed up backups. Good backup strategy is
neither simple nor cheap and is highly dependent on your business needs.
--
Mladen Gogala
Oracle DBA
http://mgogala.freehostia.com
Other related posts: