Yes, Step 4 is to CYA. Tape isn't 100% foolproof for data retention, and
the closer you get to Forever, the less stable it will be, so having two
copies would help ensure you have the data should a request be made for it
in the future.
Let's use an imaginary media called Tangwort. If you say that Tangwort has
a 90% reliability factor after 10 years on a shelf, then having two copies
of the data gives you roughly a 99% chance of having the data, or 1% chance
of the same data being corrupted in both tape sets. I'll take 99% over 90%
any day. Tape should be more reliable than Tangwort (mostly because I just
made it up - Tangwort II will be better), but the same rule applies - 99%
reliable media gives you 99.99% coverage with two copies.
So, if the data is really that important to have at some unknown future
date, you need to analyze how reliable your media is, and then what loss is
acceptable, and figure out how many versions you need to make to do it.
And yes, turning off expiration for the whole process would probably make
sense, too, so the data set doesn't change, and you don't lose any data
that may be important in some unknown number of years.
Nick Cassimatis
nickpc AT us.ibm DOT com
Today is the tomorrow of yesterday.
bbullock
<bbullock@MICRON. To: ADSM-L AT VM.MARIST
DOT EDU
COM> cc:
Sent by: "ADSM: Subject: Re: Eternal Data
retention brainstorming.....
Dist Stor
Manager"
<[email protected]
.EDU>
08/16/2002 02:03
PM
Please respond to
"ADSM: Dist Stor
Manager"
Good write up. Infact, it is what we are considering doing at this
point. The only question I have is about step 4. Why would you go through
and create another copy of everything? That would give me 2 copypool copies
of all the data. I guess you're going for a extra bit of CYA for a possible
failed tape?
I' thinking that the expire inventory would need to be turned off
during the second round of the instructions, and that would be a problem. I
think I'll get buyoff that the 1 copypool copy is all we got. If a tape
goes
bad, we did the best we could given the scope of the request and the volume
of data involved.
Thanks,
Ben
-----Original Message-----
From: Nicholas Cassimatis [mailto:nickpc AT US.IBM DOT COM]
Sent: Friday, August 16, 2002 8:46 AM
To: ADSM-L AT VM.MARIST DOT EDU
Subject: Re: Eternal Data retention brainstorming.....
What kind of shelf life are you expecting for your media? Since they've
discovered that optical has data decay, it's not even "forever"! I've seen
some others on the list point at it, but how will you restore this data in
10 years?
Here's what I'd look at doing:
1. Take about 5 dbbackups (or snapshots) and send them to the vault. Take
a few OS level backups of your server (bootable tape images). Send them,
too. Send volhist, devconfig, a prepare file, TSM server and client code,
OS install code, everything else that makes your environment your
environment, including the clients, offsite. This is DR - in the most
extreme sense of the term.
2. Box up the vault. Seal the boxes, leave them there.
3. Start marking offsite volumes as destroyed (or just delete them) in
TSM, and run more Backup Stgpools. They'll run longer, as you're
recreating the old tapes.
4. Go back to step 1 and repeat once. If this data is really that
important to have forever, make sure you can get it back!
5. Start sleeping - you're going to be WAY behind on that!
Now, for the people making the requirement - they need to get a contract to
have accessible like hardware to do the restores to. Not just the TSM
server and tape library, but the clients, too.
Having the data available is one thing, being able to restore it is
another.
Nick Cassimatis
nickpc AT us.ibm DOT com
Today is the tomorrow of yesterday.
|
