Re: Comments?

On Jul 22, 6:28 pm, "Dereck L. Dietz" <diet..._at_ameritech.net> wrote:
> The below is a portion of an email describing how the plan for disaster
> recovery has been explained to us.  This is an Oracle 10g database running
> on Windows 2003 server.
>
> The way we did the disaster recovery backup is:
>
>  Step1:
>
>   1.. Export the entire db (with no rows option) - This will get a copy of
> export to recreate the database with all users and system settings.
>   2.. Export the entire schema (with no rows option) - This will get a copy
> of export to recreate empty table shells with indexes, keys and all
> procedures, packages, functions and any other metadata for each user.
>   3.. Export every table by schema (all table data) - This is all the data.
> Step2:     Every export file is zipped and encrypted using gpg
>
> Step3:     Move the whole archive to USB drive.
>
>  The entire process takes about 5 full days. Which is ok considering its
> once a month job. Most of it is automatically done except for moving to usb
> and preparing the scripts. The total size of this is about 170GB.
>
> We have 1tb disks which can hold up to 5 or 6 of these copies.

In addition to all the other excellent advice you've received, note that rman compression works very well, so not only is it faster and able to do transactional recovery, it also doesn't need that zip step. It should be a lot less than 5 days, too.

The procedure you've described is very good - for long term archival purposes, not for disaster recovery - unless your disaster recovery is on some incompatible platform. And like Sybrand pointed out, the imp would be slower than the exp, 5-10x in my experience. So unless your disaster recovery window is on the order of months, you might be unnecessarily manufacturing another disaster. Which would be Ed's point. Even so, five days sounds long, I do 40G in about 40 minutes on middling risc hardware with a low-end SAN, 6-12 hours to exp/move/ create schema/imp on another box (I've been testing a complex migration/upgrade, have the whole thing down to 26 hours without compulsive tuning [perhaps next one less now that I've discovered the log file size advisor]). Google for Connor McDonald's site, he has a page on how to speed up exp. DIRECT=Y with the largest buffer made a big difference for me.

There could be an argument for user managed backups for disaster recovery, but again, you need to know what you are doing and need to have appropriate hardware (for example, a SAN with snapping capability). Personally, I like the idea of having everything including the oracle installation for a bare-metal rapid disaster recovery, but I don't know about Windows there (I tried it once with ghosting including an app server, it worked, but never felt good about the steps it required and didn't implement it).

jg

--
@home.com is bogus.
http://www.signonsandiego.com/uniontrib/20080723/news_1b23qcom.html