Fayd
Diamond Member
I think you are reading it correctly and interpreting it incorrectly. The black dots are annual failure rates, with error bars above and below. The probability density is the white bars behind it. There are 2 values (annual failure rate and probability) graphed on the same chart. The probability density is what we want to check, as it is a measure of the probability that a drive will fail at a given temperature point. The shorter the white bar, the lower the probability is at that temperature point. The probability density of failure at 45C is about 0.01, and .0015/.002 at 50C leading to my '6 times' comment. The 'sweet spot' you were talking about earlier would be where both values are at their lowest, or somewhere higher than 45C. It may be higher than 50C, but we don't have data to make a conclusion about temps higher than 51C where the graph stops (the error ranges also get much bigger). Best guess using that data is that it is right around 47C.
What I was saying about it possibly being skewed is that Google (or anyone) probably doesn't run that many drives at 50C+ which can lead to an uneven measure.
it would make more sense to me that probability density is essentially a histogram of the sample data. and google does talk about the chances for skew, at the extreme ends of the data the bounds for the AFR dots widen up. meaning there's not enough data to give a more exact estimate.