Migrating from X-RAID2 single redundancy to Dual redundancy

Note the publication date is 2/2007. The study is still interesting, but it would be useful if Google (or someone else) updated it to cover modern drives.

Mr. Schmidt's thoughts on drive temperature are not consistent with the paper he cites. Also he misstates their comments on SMART. They found that after after the first reallocated sector, drives had a 21x greater chance of failure within the next 60 days (compared to drives that had no such failures). This is not not about watching the drive for the first 60 days only. More generally, they only found 4 SMART parameters are "strong", and found that 36% of the drives failed with no SMART errors being reported prior to failure.

Google also clearly does not view all drive models as being equally reliable. "...Failure rates are known to be highly correlated with drive models, manufacturers, vintages..." It would be truly useful if Google (or other folks running large data centers) published their actual reliability data on the drives they use (identifying %failures by model).