[Techtalk] Running servers too hot?

[Akkana Peck]

Sarah Newman writes:
> If Maria's way doesn't work, you might try digging around in 
> /proc/acpi/thermal_zone/THRM, but I am guessing if one doesn't work the 
> other method won't.

I've found them to be pretty different. I've never actually gotten
lm_sensors to work on any of my machines, but I've had several
on which cat /proc/acpi/thermal_zone/*/temperature works
(note, it's not always THERM, but there's usually only one
directory under /proc/acpi/thermal_zone so * works).

But beware: just because it prints something that looks like
a temperature doesn't mean that it's the real temperature.
Quite a lot of manufacturers apparently write acpi code that
looks something like this:

int report_ACPI_temperature()
{
    if (os == "Win95" || os == "Win98")
        return win9xtemp();
    else if (os == "NT")
        return NTtemp();
    else if (os == "WinXP" || os == "Vista")
        return WinXPtemp();
    else
        return 32;
}

So check the temperature at idle and again in the middle of a
kernel compile or something, and if they're the same both times,
be very suspicious. Via seems to be especially bad about this;
at least, out of three Via BIOS machines we have here, two always
return the same number no matter what, and the third doesn't
seem to provide temperature via ACPI at all.

> If neither works you might try opening the case and getting an IR 
> thermometer and measuring the temp when it hangs, but probably opening 
> the case will change the heat properties too much.

Or you could try running one with the case open and a room fan
blowing on it and see if it stays up longer; but if your co-workers
don't believe temperature is the problem they might not be keen on
experiments like that.

	...Akkana