Comment 16 for bug 412152

May I point out that if Palimpsest raises false positives, it will cost people real money?
About $50 per false warning!

It will cause some people to trash their computers. It will cause many people to spend hours trying to diagnose their disks. Many people will lose irreplaceable data in the course of responding to these warnings.

This is an extremely serious bug, if bug it be.

Me? I chucked out an old disk drive and spent about 5 hours reinstalling Ubuntu. On the *second*, newer computer, I got a bit suspicious. I found a paper by Pinheiro, Weber and Barroso
at Google: http://research.google.com/archive/disk_failures.pdf

The found that "Despite [correlations between failure rates and SMART data] we conclude that models based on SMART parameters alone are unlikely to be useful for predicting individual drive failures."

Specifically, from discussions on Redhat's list, it seems that Palimpsest puts up a warning when one sectors is reallocated, but Google says "85% of drives survive more than 8 months after their first reallocation." And further, they state "Out of all failed drives, over 56% have no signal on any of the four strong SMART signals... Actual useful models, which need to have small false-positive error rates are in fact likely to do much worse than these limits might suggest."

That says it all. Palimpsest is a misbegotten piece of software that means well, but will ultimately hurt people more than it helps. Sorry, developers. Ya gotta do more than write good code: you need to pass a real statistical cost-benefit analysis, and the inital results suggest that it's a bad idea.