On Tue, Mar 2, 2010 at 8:06 AM, <and...@msu.edu> wrote: ... > I've seen the smart system report errors and have had them become > true a few times, but far more often I've seen the damn things report > "No proble, Boss" and then died a little later...
I seem to recall a USENIX paper from google (perhaps for the FAST conference?) in which they analyzed the failure statistics for their server farms against things like the SMART stats and a bunch of other stats they collected. IIRC, they found SMART error reports to be useful in predicting failure and that while there were some correlations in other stats, the false positive rates would make using them uneconomical. Or something like that. If you're concerned about this disk failure then you should hunt up and read the actual paper...and continue your research beyond that. Philip Guenther
