@Joe Kraska

You can't take a MTBF and divide it by the number of hours in a year to get the proportion of drives that fail in a year.

You need to take into account the distribution of the failures, and it's most certainly not flat.

http://research.google.com/archive/disk_failures.pdf
"Or basically forever".

That's one way of looking at it. The other way;

In a lot of 159 of these, 1 will fail annually.

C//
@Joe Kraska

You can't take a MTBF and divide it by the number of hours in a year to get the proportion of drives that fail in a year.

You need to take into account the distribution of the failures, and it's most certainly not flat.

http://research.google.com/archive/disk_failures.pdf
"Or basically forever".

That's one way of looking at it. The other way;

In a lot of 159 of these, 1 will fail annually.

C//