Disk failures in the real world: What does an MTTF of 1,000,000 hours mean to you?

In large-scale systems where number of components can approach a million a failure is a significant problem. In this paper authors have presented and analyzed failure data from different large systems. More than 100,000 disks with different interfaces which come from at least four different vendors have been investigated. According to information which one can find in datasheets the annual failure rate should be at most 0.88%, but the results show that replacement rate is usually in the range of 2 to 4% and can be up to 13%. Another important outcome is that infant mortality is negligible, but wear-out starts earlier than expected and replacement rate is constantly increasing over the time. For different disk interfaces replacement rate is at the same level. Exponential distribution does not provide good modeling for time between