| 1 |
On Wed, Nov 12, 2014 at 7:02 PM, Volker Armin Hemmann |
| 2 |
<volkerarmin@××××××××××.com> wrote: |
| 3 |
> Am 13.11.2014 um 01:01 schrieb Adam Carter: |
| 4 |
> |
| 5 |
> "Backblaze's analysis of nearly 40,000 drives showed five SMART metrics that |
| 6 |
> correlate strongly with impending disk drive failure: |
| 7 |
> |
| 8 |
> SMART 5 - Reallocated_Sector_Count. |
| 9 |
> SMART 187 - Reported_Uncorrectable_Errors. |
| 10 |
> SMART 188 - Command_Timeout. |
| 11 |
> SMART 197 - Current_Pending_Sector_Count. |
| 12 |
> SMART 198 - Offline_Uncorrectable" |
| 13 |
> |
| 14 |
> http://www.computerworld.com/article/2846009/the-5-smart-stats-that-actually-predict-hard-drive-failure.html |
| 15 |
> |
| 16 |
> |
| 17 |
> everybody with half a brain would figure that one out themselves. ... |
| 18 |
|
| 19 |
It is still useful data nonetheless. The fact that other factors were |
| 20 |
not correlated is interesting in and of itself. The difference |
| 21 |
between a good hypothesis and a good theory is data. |
| 22 |
|
| 23 |
There aren't a lot of published accounts of drive reliability across |
| 24 |
large sample sizes of any kind, so this stuff is useful. |
| 25 |
|
| 26 |
Sure, it would be really nice if somebody bought a million drives at |
| 27 |
random, removed the labels and modified the internal vendor |
| 28 |
identifiers, distributed them to three independent academic testing |
| 29 |
labs, and performed a series of double-blind tests for a span of 5 |
| 30 |
years. But, since nobody seems willing to just spend that kind of |
| 31 |
money for the good of mankind, we can at least read accounts by |
| 32 |
companies like Google/Backblaze/etc when they publish them. |
| 33 |
|
| 34 |
-- |
| 35 |
Rich |
Archives