Difference between revisions of "Miscellaneous/hdd-lifetime"
| Line 45: | Line 45: | ||
== Conclusion == | == Conclusion == | ||
Hopefully you've now realized that this is all statistical data, only really | Hopefully you've now realized that this is all statistical data, and is only really meaningfull if you have a big array of disks. It doesn't apply very well (with great level of certainty) to your single datapoint (your bubba).<br> | ||
<br> | <br> | ||
So, all you can really do is to remember to always backup data you care about (pictures, docs etc). Rule of thumb I use is to have them on three places, and all of them can't be in the same | So, all you can really do is to remember to always backup data you care about (pictures, docs, hg repos etc). Rule of thumb I use is to have them on three places, and all of them can't be in the same geographical spot (not all in same closet in the same house!) :-) | ||
<br> | <br> | ||
Revision as of 08:06, 21 October 2011
How long will my Bubba HDD last?
So, your bubba have been running for a couple of years, how long will it last?
There is not easy answer to that question :(
There are alot of metrics around, MTBF, CDL, AFR
WD (excito's main HDD vendor) used to quote 30,000 hours of MTBF. But all that really means is that if you had a large sample of drives, half of them would be broken after 3.4 years.
Another example, using Googles "real world" AFR data (link 2 below), the AFR for the different disk ages:
| Disk Age (years) | AFR |
|---|---|
| 1 | 1.7% |
| 2 | 7.9% |
| 3 | 8.6% |
| 4 | 6.1% |
| 5 | 7.0% |
So if you have a server room with 1000 disk, you could expect 1.7% (or a total of 17) of them to be broken after year, a total of 96 after 2 years etc...
Conclusion
Hopefully you've now realized that this is all statistical data, and is only really meaningfull if you have a big array of disks. It doesn't apply very well (with great level of certainty) to your single datapoint (your bubba).
So, all you can really do is to remember to always backup data you care about (pictures, docs, hg repos etc). Rule of thumb I use is to have them on three places, and all of them can't be in the same geographical spot (not all in same closet in the same house!) :-)
Some links
- WC Official Statement
- Google's Experiences of BIG disk arrays
- Why big MTBF numbers are pointless
- More on MTBF