How often do servers really fail? If Backblaze’s 61,523 operational hard drives are any indication, not often.
On average, Backblaze reports its drives have a fail rate of 1.84 percent in Q1 2016. That’s 266 failures with over 5 million days in operation.
Cumulatively, Backblaze drives have over one billion hours of operation to date. Its most used drive was a 4TB Seagate, while the 2 and 3TB HGST have a slightly better lifespan.
Western Digital drives failed the most, while Seagate and Toshiba have roughly the same drop-dead ratio. HGST was far and away the best drive on Backblaze’s test; it routinely hovers around one percent failure rate.
Backblaze notes it uses three components when considering failure: a drive that won’t spin or connect to the OS, one that will not sync (or remain synced) to a RAID as well as some in-house analytics.
Backblaze obviously isn’t the only horse in the race when considering servers, but it’s a good look at best practices for those interested in building their own server. If your app or service depend on it, looks like HGST may be the way to go.