ADSM-L

Re: Determining which side of an optical platter a file is on?

2005-10-15 13:59:08
Subject: Re: Determining which side of an optical platter a file is on?
From: "Allen S. Rout" <asr AT UFL DOT EDU>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Sat, 15 Oct 2005 13:58:52 -0400
==> On Sat, 15 Oct 2005 12:57:51 -0400, Justin Derrick <ADSM AT JUSTINDERRICK 
DOT COM> said:

> I'm figuring that doing a query that simply tells me which side of
> the platter the data is on, or which offset from the beginning of the
> disk (and determining which side it's on by saying a number is
> greater than or less than a particular block that represents the
> half-way mark) will enable me to do things that much faster.

All you need in order to save this effort is enough disk to do -one- volume at
a time.

-
Start with empty DISKpool

MOVE DATA optical-vol tostg=DISKpool

Then you can q contents of everything on the diskpool, and process to your
hearts' delight.

Then empty the diskpool to whatever the next home is of this data.

rinse, repeat.
-

One mount per optical volume (well, in fact probably two or three; one file
bridging the beginning, one file bridging the end) and you don't have to worry
about which side whatever is on.

> I'm trying to avoid a scenario where I have to make a backup
> storagepool, for several TB of data, then mark platters as being
> destroyed in order to make retrievals efficient, then copying that as
> I perform my extraction.  I simply don't have that much spare disk
> lying around.  =)

If you're nervous about the integrity of the data, but you're not, say, $50K
nervous about it, then you're not that nervous.  So don't be nervous. :)

That $50K would get you plenty of cheap SATA raid to hold ALL of your optical
stuff.  $10K would get you enough cheap SATA raid to make your transition
drastically easier, and keep you at one mount per volume. (you could do the
above process with 100 or more volumes at a time).

With 2000 volumes and lots of babying, it's reasonable to talk about spending
$10-20K on hardware, vs. spending $10-20K of your time fiddling with it.  At
the end of the former, you've still got 3-4TB of disk.  At the end of the
latter, you've got an exasperated admin. :)

You do have a backup of this stuff already, right?  If not, please please do
that first.  I've found 3995s to be the wost, in terms of data reliability, of
any IBM media I've used.  We ditched them entirely.

For everything we used to do on optical, we're now using disk of one
description or another.  Very Slow SATA raids can probably serve your
retrieval needs, and provide what will appear a HUGE performance (latency)
improvement to the end customers.

In any case, if you don't have a stgpool backup of it already, than make one.
Then you're back in the 'convenience and efficiency' domain of optimization,
instead of the 'Will I lose data by mounting this platter' domain.


- ALlen S. Rout