18 Mar
2005
18 Mar
'05
1:23 a.m.
[a bunch of replies all in one]
On Wed, 16 Mar 2005, Vintage Computer Festival wrote:
My assertion is that it's a false economy; besides bit-rot,
there's future (near-guarenteed) incompatibilities, and if enough
time passes in multiple media (papertape... 1/2" mag tape...
DC300... 3740... QIC... etc)
In fact, with Moore's Law and related continuous hard disk is *far
cheaper*. It's just that you're not deferring the cost to your
future inlaws!
No point in insuring the data contained in the disk drives in the
computer in the back seat; it's just a copy.
On Thu, 17 Mar 2005, Doc Shipley wrote:
However, I find that media is much more sturdy than
these
discussions would indicate. We tend to gripe very loudly (and rightly so)
when poor quality backup media takes our data with it. But what about all
those times we go back to our old backups and thankfully find what we
need?
That's true. It's not like 100% of all media rots to nothing. But
"finding" old media that's still readable is different from
"archiving".
I consistently read lots of data with nary a problem:
20-30 year old
floppy disks (5.25" and 8"), 20-30 year old mag tapes, even 20 year old
VHS tapes. And of course the punch cards don't count ;)
No offense, but you're in the business of doing so :-) Only a
minority of even us can read and convert old formats, even when
readable.
On Thu, 17 Mar 2005, Jim Leonard wrote:
If you somehow
think that tapes stored in a controlled vault is
more reliable, or less susceptible to bit rot than rotating
spindles, I believe you are wrong (and my every experience and
observation says otherwise). Only fiche and paper are statically
reliable.
You are omitting cost. Tapes in a vault cost significantly less over 10
years than the cost of 4 live systems + replacement hard drives + network
bandwidth + electricity for your method. Your method, unless I am misunderstanding you, has no
revisioning. It
protects against hardware failure, but how many revisions do you keep? What
protects against you mistakenly deleting the wrong directory, etc.?
That's what filesystems are for.
I also extended a rotating backup scheme I found on the net, that
uses hard links and an intentional feature of rsync; it fully
saves N copies of the data with true incremental changes with only
about 10% more diskspace than the original files themselves, for
data that changes at a "reasonable" rate of change (per week,
say). (Example: I have a big RAID server that has 5 weeks of 100%
snapshots; the original data is about 360GB, 5 weeks worth is
about 500GB.
http://wps.com/temp/rotate-backup
I agree completely that archival data should be
monitored in some fashion
instead of being locked away forever. And, cost permitting, transferred to
new media.
On rotating spindles, this is continuous...
On Thu, 17 Mar 2005, Dwight K. Elvey wrote:
One other thought is that one should have a way to
transfer the raw data from what ever media you have.
This allows one to have redundent information stored
on the same media. If you depend on normal file systems,
you run the risk of the file system not allowing you
to access the data simply because a small part is
damaged.
Hence multiple, ordinary servers, physically separate. The chances
of all of them developing a bad disk simultaneously is exceedingly
low.
If one of my servers craps a disk, that machine is repaired; data
from the other server(s) is automatically (cron) copied
to it.
Self-healing.
On Thu, 17 Mar 2005, Doc Shipley wrote:
>>> This reminds me that I read not too long
ago that many of the super
>>> computer labs ship PCs between sites because it's *faster* to ship a
>>> working PC with 1TB of disk containing data than it is to transfer it
Yikes, brings
new meaning to the term "Our System Crashed." Wonder
what the insurance is on something like that....
The insurance in that case was basically irrelevant. Their insurance would
cover the hardware replacement costs, period. Then the other issue was the archived files
themselves. Some of it was in
PageMaker, some in Wordstar, some in MS-DOS .prn files, and a lot of it in
AutoCAD r12 and a freeware DOS GIS tool whose name I've mercifully forgotten.
THen there's that! Added to the top of media recovery ("who's got
a 200bpi UNIVAC drive?") it generally means that the data is
simply "lost". (Few think to "SAVE-AS" plain ASCII for posterity;
I rarely do.)
I spent several hundred hours sorting backup sets,
restoring them to disk,
migrating files into current formats, and then archiving both current and
original copies to 4mm tape. It made for some very soothing afternoons,
hidden in my office.
And they only did it because it was required by law!
I'm seriously planning on printing out my website on paper. Laugh,
but it might be one of the few copies that survives long term.
That is, if I can find a printer that produces real output.