Opinions on safe disk temperatures

[John Doe]

i can recommend the silverstone 180mm AirPenetrators!
available with 4 pin or 3 pin.
2 of those fans fit quite good in the front of a midi tower.
connect these fans to a corsair commander pro*1
to get it even more silent, sharkoon launched the vibe fixer several years ago.

this combination is for me as a noise sensitive person really good. the server is 110cm away from my ears and i am not geting crazy.


*1
this corsair commander pro is nothing worth to recommend. it is basically nothing else like a temperature regulation, not just control.
it gets shipped with 4 temp sensors and you can put fan curves on it. after set up, just connect it to power, no usb required.
that is unfortunately the only product i know with this functionality. but there are drawbacks. eventhough it is advertised as "pro" you cannot put smart logic in it, like if sensor 1 or sensor 2 is xy, then fan 1 and fan 2 xx rpm, else fan 1 yy rpm.
also grouping is not possible via software, so you cannot group fans.

 

[-fun-]

OP is likely using the new 8TB Reds which use air inside, unlike the older helium modesl [...]

Yes, these are very new 8TB models, the 3TB disks are >2 yrs old. Finally an explanation for what i observe, thank you! On a side note: My new 6TB disks I recently returned (SMR ...) were not nearly as hot.

 

[Kevin Horton]

I will do whatever it takes to keep my disks below 40°C, but it is hard to know how important that really is.

I have one 4TB Seagate ST4000DM000-1F2168 that I shucked from a USB enclosure. That drive spent the first 26,000+ hours of its life living in that enclosure. Now that it is out of the enclosure, I can read the SMART data, and it shows a lifetime maximum temperature of 59°C. I have no way to know how much of its life was spent running in the upper 50s, but it was in a room that didn't have extreme temperature variation, and the air vents in the enclosure were not blocked, so there is no reason to believe that the temperature varied too much. It is quite likely that it spent most of its life running very hot. The disk shows no obvious sign of the abuse - all the SMART data looks completely normal.

Maybe we worry too much about disk temperatures.

 

[John Doe]

https://static.googleusercontent.com/media/research.google.com/en//archive/disk_failures.pdf

 

[Constantin]

Your sample size is too small. It’s like claiming that the centennial light bulb is emblematic of all incandescent light bulbs and therefore all incandescent light bulbs will last 100 years.

It reminds me of some pointing to aspects of backblaze research discounting temperature impacts without noticing that the vast number of disks at Backblaze never went over 31*C. Some seagate models had failure rates that trended noticeably upwards starting at just 26*C.

So I’d stick to large datasets with disks operating at 40*C and up rather than extrapolating from a very small sample or looking at disks operating near perfect operating temperatures. The devil is in the details.

 

[Constantin]

The crazy thing about this topic is there are contradictory results from large datasets. For example, this paper presented at USENIX in 2016 looked into the total operating costs associated with data centers. The authors found that temperature wasn't a great cause of failure in their dataset, but humidity was.

The paper helps explain the trend towards free-cooled data centers vs. the traditional chiller / CRAC approach. The authors acknowledge in Table 3 that the annual failure rates are double in a free-cooled environment vs. traditional designs. However, the added costs associated with doubled AFRs are more than offset by two factors:

• lower operational and CAPEX costs for free-cooled data center designs
• the refresh rate associated with HDDs in data centers. The MS data center seems to be replacing them on a 4-year basis, which may or may not be representative of your use case.

Unfortunately, the whereabouts, operational conditions, disk drive models, etc. are kept under wraps. So it is hard to say anything about this data other than it is applicable at Microsofts data centers based on how they use them. However, I expect MS and other data center operators to increasingly build free-cooling data centers in cool, dry places that feature cheap power (i.e. hydroelectric dams) and then trench a couple of fiber optic cables to them.