If you wanted to talk operating systems, I know that part. I’m just not familiar enough with the verbiage about hard drive failures. The reason I’m here is to find out which models or brands to stay away from.

Indeed, the Dell BOSS is essentially a RAID card that supports one m.2 drive, or two m.2 drives in RAID-0 or RAID-1. I’m kinda surprised it provides any SMART access at all.

I hope we can live up to your expectations. Thanks.

We are trying to decipher how each SSD records various SMART stats things like total block written and read, etc. There is little consistency between manufacturers on what they report and what they name the SMART attributes they using. We’ll keep at it, as what you are asking for would indeed be interesting.

Re temperature variance: I have definitely observed this. Example: Dell R730xd with a rear cage. The rear cage holds 2 (or 4?) drives that we used for boot, so that front-panel bays were available for capacity drives. The rear bay / boot drives consistently reported temperatures several degrees C higher than front-panel drives. TFA mentions distance from the cold aisle, which could be a factor, as well as general differences in internal airflow. SSD temperature is also a function in part of where on the drive the sensor is located on the device – some parts generate more heat and get more airflow than others. Temperature is also a function of workload – an idle drive generates less heat than one under heavy write load. Boot drives are also typically lower in capacity, which could mean more room for internal heat dissipation, but also that they generate less heat than a larger, busy capacity drive. Which could mean that the delta due to location is actually higher than the net measurement.

Re Samsung:
There are other brands that are not represented much or at all here. One factor is whether a given brand even sells boot drives – which are not always identical SKUs to client / desktop.

Another is that they speak of boot drives coming bundled with chassis. In such a case, the choice is often up to the chassis vendor. With Dell, for example, sometimes one can choose a specific drive model, but they also have SKUs marked as AG for agnostic, meaning that they may have multiple manufacturers or models that they use as they see fit to fulfill a given order. This is done to shield the end user from supply chain vagaries including lead times and products going EOL. So likely BB is not explicitly avoiding Samsung, Micron, Kingston, Kioxia, Intel/Solidigm, or anyone else.

In the case of the Dell BOSS, since we see the term VD, it’s possible that the card has an RoC used to present a Virtual Drive instead of passing the physical drives through (which may also be an option). The VD layer may not surface isolated write errors on the physical drives to the system, i.e., it may map around them transparently, or it’s possible that they aren’t monitoring for a VD with one mirror failed. So the stats for those may not be congruent with those for the other, directly-managed drives. The drives/sticks on the BOSS card are not proprietary.

They’re overkill for the pod boot drives since all those need is something to boot off, but we do use them for some other things. We don’t currently have good publishable drive stats on those, though. Partly because the stats collection is designed around the pods, partly because there aren’t very many of them compared to the number of pods, and partly because some of them are cache drives that get worn to failure well past their rated endurance.