[coyotos-dev] RAID sucks

[Jonathan S. Shapiro]

On Mon, 2007-07-23 at 07:09 -0700, Eric Northup wrote:
> Actually, as I understand the recent large-scale disk drive reliability 
> studies (two were presented at a recent FASTA), drive failure was *not* 
> strongly correlated with temperature within their operating temperature 
> range.

Umm. But that's just the point. In this case the SMART data is clear
that both drives exceeded operating temperature range multiple times.
They were stacked vertically within in a 3 to 4 drive frame. Either the
drives themselves went, or heat rising from the lower two drives drove
the temp on the upper drives above threshold.

I also suspect (without any evidence at all) that things are a bit
different in large data centers. I'm stuck in the "unserved middle" at
the moment. I know enough to know that heat dissipation is important,
and I run a controlled environment, but I don't necessarily know how to
configure the case fans for proper airflow in this case.

> The 
> strongest predictor of failure was to have another drive with the same 
> manufacturer/model/lot fail (ie, don't buy matching pairs), and that 
> does match my experience - and yours, it sounds :-(

And of course, all of our drives were from the same lot. Even so,
correlated failure within 3 hours seems unexpected. This is why I am
inclined to believe the "heat rises" theory first.

> I don't know how much this matters to the overall failure rate, but 
> there is a significant loss in how much (and how affordably) data can be 
> recovered once you start using disks with >1 platter (such disks also 
> produce more heat).

Interesting. Most over-the-counter drives these days have multiple
platters.

For personal systems and small businesses I'm really glad those rescue
folks are out there. Somewhere around 4-5 people the cost of the
redundant backup server becomes justified, and after that you shouldn't
need them.


shap