On the usefulness of S.M.A.R.T.

Rec 28-fev-2007 12:37

I read this nice paper from some folks at Google where they discuss failure trends in large hard disk populations. One of their most interesting conclusions is that S.M.A.R.T. parameters don't seem to be good predictors of individual drive failures -- more than half their failed drives had their S.M.A.R.T. parameters just fine, even though the other more-or-less half did in fact showed anomalies.

This doesn't mean S.M.A.R.T. is useless. It seems fine for statistically studying a large population. It is just not good for predicting which drives are about to die.

I particularly liked their pragmatic definition of a failed drive: one that had to be replaced.

The study was specially credible because the population of drives they analyse is quite substatial, in the order of hundreds of thousands.

But I won't spoil your reading any longer. Get the paper here.
top