Perishable Media

CDs are perishable media. Hard disks are not. The current trend is to use large hard disks and networks to replace any need for CDs. An emerging trend is to use RAIDs to make hard disk storage persist as long as necessary. RAIDs can be configured to have mean lifetimes of hundreds of years. Thousands even, but nobody really wants that.


Do you have any studies to back up that assertion? I've been told that the average MTBF for a modern hard drive is only 5 years -- not much longer than the warranty period. I vaguely remember reading that average-quality CDs should be readable for 50 years. How would anyone know that, when CDs haven't been around for 50 years?

I agree that using hard drives configured in a RAID should last much longer than a collection of CDs that are never copied. (Distributed mirroring over several continents should last even longer). But wouldn't using RAID concepts work just as well with CDs as they do with hard drives? In other words, use enough "extra" CDs so when one fails, you can reconstruct the data from the other ones in your collection. Periodically scan the entire collection of CDs, and *when* one fails, reconstruct a fresh new copy.

(I wish I had a pointer to failure analysis studies here. assumes MTTF = 100 000h = 11.4 years for individual disks ...).

 MTTF(powered-on-hours): 1.2 M  20 K  8.8 K
for 3 different hard drives -- for an always-spinning computer, that translates to
 MTTF(powered-on-years): 136 years  2.3 years  1 year
-- what a wide range of times ! ... but of course few computers, and even fewer laptop computers, keep their hard drives spinning 24/7. )

SimplestThingThatCouldPossiblyWork: simple mirroring, each data disk backed up by a mirror disk (RAID1). If any 1 disk fails, all data can be recovered from its mirror. More complex schemes can protect 4 disks worth of data with 4 "parity" disks such that if any 2 disks are destroyed, all the data can be recovered.

I suppose you could do "mirroring" with paper and a copy machine as well, come to think of it.

-- DavidCary

What's important isn't how many disks are used for parity. What's important is how many dimensions of parity there are. For every additional dimension in the array, the MTTF gets multiplied by a factor of a hundred, as long as the total number of disks stays constant. It's very easy to build a RAID with a MTTF of a million years, with extremely reasonable overhead.

And the reason why CDs, paper and microfilm are so perishable is because they aren't live media. As a consequence, it's impossible to perform the error-checking that's necessary to take advantage of any mirroring scheme. Hard disks function in RAIDs because you know when one fails.

On a related topic, there are now media which are even more perishable -- such as DVDs with a special coating which causes them to become unreadable a few days after being exposed to air. (While in a sealed container; they last a long time). Used for a returnless movie "rental" system -- you pay a small fee, receive the disk, can watch it as often as you want until the disk fails; then throw it away.

There's more than one dimension to perishability, though. Those CDs might still be intact after 50 years, but without a working CD player they are just pretty drink coasters. Part of the company I work for is in the business of large scale archival document management for UK government departments. They have returned to long-term storage on microfiche, since properly prepared and stored photographic media have indefinitely long useful lifetimes, and all that is required to recover the information is a magnifying glass.

View edit of November 27, 2014 or FindPage with title or text search