2 ms·

The statistics of small sample sizes comes into play here. The small sample size of a personal environment doesn't lend itself well to comparisons to a large sa

The statistics of small sample sizes comes into play here. The small sample size of a personal environment doesn't lend itself well to comparisons to a large sample size of a cloud storage provider. For example, their 2% failure rate of Vendor X drives doesn't mean anything to you if you don't have enough drives. Your few drives might be in the 2% or the 98%. The 2% failure rate is only applicable to the population of drives as a total. It doesn't tell you anything about your drives in particular. Now, on top of that, their usage patterns don't necessarily reflect what a typical user would see on their personal computer. I suspect that their read/write patterns skew towards the write-once, read-rarely category. Even for a typical enterprise datacenter, their usage patterns are going to be different.

arrrg 7y ago
I’m not really sure what you statistical point there is. If Vendor A has a 2 percent failure rate and Vendor B has a 3 percent failure rate it’s still valid to conclude that you would be less likely to have a failure with Vendor A even if you only ever buy one drive. Home use being not comparable to their usage patterns is a valid argument, but your statistical argument seems like nonsense to me.
- mbreese 7y ago
  Population level statistics aren’t informative at the individual sample level. If you buy 100 hard drives, a 2% failure rate makes sense and you can reason what it means. You’d expect two of those drives to fail in a year. But what does that mean if you only buy two drives total? An individual drive can’t fail by 2%. It has either failed or it hasn’t (ignoring bad blocks, etc). So, for home use, the Backblaze statistics are interesting, but can you really tell the difference between a 2% failure rate and a 3% failure rate when you buy a small number or drives (or likely just one)? You’re most likely to not see any failures from either vendor, as they have a success rate of either 97 or 98%. It’s only when your sample sizes get sufficiently large that these failure rates start to have a good predictive value.

arrrg 7y ago

I’m not really sure what you statistical point there is. If Vendor A has a 2 percent failure rate and Vendor B has a 3 percent failure rate it’s still valid to conclude that you would be less likely to have a failure with Vendor A even if you only ever buy one drive. Home use being not comparable to their usage patterns is a valid argument, but your statistical argument seems like nonsense to me.

mbreese 7y ago
Population level statistics aren’t informative at the individual sample level. If you buy 100 hard drives, a 2% failure rate makes sense and you can reason what it means. You’d expect two of those drives to fail in a year. But what does that mean if you only buy two drives total? An individual drive can’t fail by 2%. It has either failed or it hasn’t (ignoring bad blocks, etc). So, for home use, the Backblaze statistics are interesting, but can you really tell the difference between a 2% failure rate and a 3% failure rate when you buy a small number or drives (or likely just one)? You’re most likely to not see any failures from either vendor, as they have a success rate of either 97 or 98%. It’s only when your sample sizes get sufficiently large that these failure rates start to have a good predictive value.