3 ms·
Disk failures in the real world: What does an MTTF of 1M hours mean to you?
- fn-mote 4y ago(2007) Not that it's not interesting, but ... I wonder what the last 1.5 decades have brought in improvements.
- hinkley 4y agoI recall trying to work out how on earth we ended up with three drive failures in a 10 drive array literally out of the box and then I started doing some statistics on the premise that 1M MTTF means 1/2M chance of failure per hour. This is the wrong math not but it still illustrates how fast failure rates ramp when you multiply by the number of opportunities. When you have 10 or 20 hard drives you start seeing failures in months to years. If you have a thousand drives you’ve basically got a full time job replacing drives. And that’s why you have companies making or designing systems where failing drives are shut down and maintenance is only performed every 3-4 failures. And once in a while ones where you just decommission the entire rack when the faults goes over a threshold.
- hinkley 4y agoI recall trying to work out how on earth we ended up with three drive failures in a 10 drive array literally out of the box and then I started doing some statistics on the premise that 1M MTTF means 1/2M chance of failure per hour. This is the wrong math, but it still illustrates how fast failure rates ramp when you multiply by the number of opportunities. When you have 10 or 20 hard drives you start seeing failures in months to years. If you have a thousand drives you’ve basically got a full time job replacing drives. And that’s why you have companies making or designing systems where failing drives are shut down and maintenance is only performed every 3-4 failures. And once in a while ones where you just decommission the entire rack when the faults goes over a threshold.