Google published failure statistics on 100,000 disks, but it’s impossible to predict exactly when your disk will fail.
Disks fail. The question isn’t, “Will it fail?”. The question is, “When will it fail?”. That’s a fact of life.
I recently had a new client express surprise when I mentioned that his computers’ hard drives will fail, sooner or later.
Google studied 100,000 of their hard drives, and in 2007, in Failure Trends in a Large Disk Drive Population, they published their failure statistics and conclusions. Google uses low-cost drives just like the kind that are in our desktop PCs. (What exactly is inside a Google data center?)
StorageMojo, in Google’s Disk Failure Experience by Robin Harris, did a great job of summarizing the results. The article points out that when a disk’s SMART reports
- scan errors
- reallocation count
- offline reallocation
- probational count
the disk is failing and should be replaced before it dies completely.
For most users, the most relevant conclusion is that a disk is more likely to fail as its power-on time reaches 3 years and more. I usually recommend replacing working disks when they have 40,000 hours or more of power-on time. (There are 8760 hours in one year.) As usual, your mileage may vary.
- Use your SMARTs to avoid disk disasters
- Monitor disk’s SMART data within Windows
- Premature hard drive failures