.
That's a very fascinating paper. Thank you, Lupin.
A couple of things, though. The paper states pretty specifically that in the mid-range temperatures (~36°C to ~47°C), that temperature is the lowest factor for failure rates. Meaning, "something else" went wrong at a far higher factor for drives maintained in those average ranges. That's not exactly the same thing as saying those are the "best" temperatures.
The observed failures at higher and lower temperatures may be independent of the temperatures, with temperature possibly being associated with other environmental factors.
Still, a very powerful argument that ~30°C and under, as well as over 45°C, are not ideal ranges by a significant statistic. And, over 45°C clearly and rapidly trends into a higher failure zone ... so that does look like a significant range to avoid.
What's interesting is that all drives from 30°C to 45°C seem to have the lowest failures as a group when 2 and 3 years old; 40°C to 45°C and up still looks significantly more failure-prone when drives get out to 3 & 4 years & up. The failures at earlier ages may be an "infant mortality" factor of new drives burning in. Once they get past the 3, 6 and 12 month stage, this temperature range appears to be the sweet spot.
The variation between 30°C and 45°C is ~0.1% between 36°C and 46°C, with it maybe expanding to ~0.3% if 30°C is set as the low end. What is scarey is to see 25°C and cooler get such a spike. The only reason I've not been doing that is the expense and elaborate configuration required to really chill drive down below that range.
Conclusion: I am not going to try so hard to get my drives under 30°C anymore. I'll be perfectly comfortable at 35°C and under 40°C. And, it will, statistically, be unlikely to make any difference over my habits for the past 20 years.

My coolest drives tend to be ~28°C-30°C, and my warmer drives tend to be ~35°C-37°C, so that paper made me feel pretty good about my habits.
I hope there's a follow-up for more modern drives on a regular basis, though. That's a 2007 paper analyzing drives installed 2001 and after during 2005-2006 time period. A good number were PATA, mixed with SATA, ranging from 80 GB to 400 GB. My smallest drive now is a 750 GB, with four 2 TB drives. I've retired all my PATA drives over the past six months and only use them for experimental machines.
One exception is a 10 GB Seagate that I'm using as the system drive on the very machine I'm posting this from; been on 24-7 as a torrenting rig for most of this year. Installed as original equipment in 2002, running at 39°C currently ... all it's SMART statistics look like the thing is barely out of the factory right now. I ran it through a rigorous burn-in when I re-installed WinXP back in June and it's been pretty cheery & dead silent before & after.
*
knockswood*
