[lug] Testing hardware

[Rob Nagler]

On 11/17/07, Jeffrey Haemer wrote:
> Standalone, automatic box, though I'm interested in what sorts of hardware
> tests you recommend for different things.  Might as well get more broadly
> educated.

I did very little testing with the first Freikometer except the usual
"burn in" and
constant software testing.  There was only one problem with the first
motherboard:
low temp.  This year I got lucky and was able to do "deep freeze" testing.

This is pretty important to do at both the low and high ends.  High temp testing
is easy enough to do (you have an oven), but low end is hard.   UNAVCO (local
company) loaned me use of their freezer/oven-thing, which goes from -70C to
+100C (or higher?).

For my test this year, I did 24 hours at 0C, and then dropped it to
-10C the next
day after I saw it cycled well the first day.  And, YAY!, it survived
at -10C.   The
blue ice I am using for thermal mass was frozen, but the device operated just
fine.

If this were a commercial product, I would test a random selection of devices.
I would also burn-in test all of the first devices I ship.

If the device has a flash device, make sure you are using an "industrial" flash.
I have had problems with flash devices dying at high and low temps.  The
industrial devices are more robust.  However, I'd still cycle the
writes and reads
through the flash drives on their own.

If you are using hard disks, beat them up.  Drop the device, etc.   There are
shake testers.

There are lots of other issues.  What types of devices might the user connect
to your device?  Is dirt/dust an issue?  Ventilation?  ...

Oh, and expect field failures that you can't figure out. :-)
NetEqualizer had this
kernel lockup only in certain environmental conditions, and it was related to a
chip bug on the motherboard.   It was very hard to isolate the problem, because
the software was doing so much it was unclear if it was a software or hardware
problem.

Rob