LM_sensors: explain 'hysteresis' in this context.

GrapefruiTgirl · 05-08-2007, 09:25 AM

Hysteresis (noun) - 'The lag betwwen cause and effect' basically.

So, I interpret this, in this context of motherboard temperature sensors, to mean something like this:

Let's say a CPU's temperature is 50.0'C and the sensor monitoring it gives a reading of 50.0'C. Ok, fine.
Now, lets say the CPU temperature increases instantly (hypothetically) to 60.0'C and it takes 15 seconds for the sensor monitoring it to also read 60.0'C. This (interpreted literally) means to me a 'hysteresis' value of 15.

So. Why is it, with for example the LM_sensors utility, would I have something like the following:

Current CPU Temp: 61.5'C (example)

CPU temp limit: 60 (degrees 'C)
hysteresis: 50 (degrees 'C)

My question is, how does the 'hysteresis' value figure into the detection/reporting of any given sensor signal? What does the hysteresis number mean in the LM_sensors configuration settings?

Over the last number of months, I've read probably a hundred LKML threads, all the LM_sensors documentation, the LM_sensors mailing-list, the docs for my onboard sensors, as well as the manufacturers specs of my chipset and motherboard, and nothing anywhere specifies exactly what the hysteresis number does or means in this context.

Explanations welcome

lazlow · 05-08-2007, 09:55 AM

GFgirl

First there is nothing that can be done to change sensor lag. It is just like starting a car on a cold morning. The engine has to warm up before you get heat to the cab.

This is my understanding of what is going on (it could be wrong). The hysteresis number are used to generate the warnings. We know how many seconds it takes for the cpu to go from hot to too hot. Using the hysteresis number we can translate the time into a temperature. So now we know the difference in temperature between too hot and just hot(say 58 and 62). So we set the warning to go off at 58, a low enough temperature to give us enough time (sensor lag) before the cpu maxes out at 62.

I hope that made some sense.

Lazlow

GrapefruiTgirl · 05-08-2007, 10:07 AM

Hi Lazlow, thank you, I understand your explanation perfectly. However, it seems from the output below, that it is not correct in the context of LM_sensors:

Code:

System Temp:      +32°C    (high =   +60°C, hyst =   +50°C)   sensor = thermistor
Northbridge Temp: +37.0°C  (high =   +60°C, hyst =   +50°C)   sensor = diode
CPU Temp:         +59.0°C  (high =   +63°C, hyst =   +50°C)   sensor = thermistor

Using the description you gave, I should be seeing an ALARM on my CPU Temp, because the reading is well above the hysteresis level.
Ideas?

GrapefruiTgirl · 05-08-2007, 10:36 AM

# Think of tempx_over as 'alarm set' and tempx_hyst as 'alarm clear'
# thresholds. In most cases the 'over' value should be higher than
# the 'hyst' value by several degrees.

Hmm... OK, the above is from the example sensors.conf file.. According to this, the _over_ value is as one would think the ALARM trigger level, and the _hysteresis_ value is the ALARM CLEAR level.
According to a science website, hysteresis is a RANGE between two values. In the case here of LM_sensors, it would appear that the _over_ value is actually the upper hysteresis number, and the _hyst_ value is the lower hysteresis number..
Let's try that

and update soon.

OK.. SO:

Code:

    label temp1   "System Temp"
	set temp1_over 40
	set temp1_hyst 36
    label temp2   "Northbridge Temp"
	set temp2_over 55
	set temp2_hyst 50
    label temp3   "CPU Temp"
	set temp3_over 65
	set temp3_hyst 62

Using temp1 as an axample, based on tests I just did, the ALARM would come active if the temperature hit 40, and the alarm clears when the temperature drops BELOW the hysteresis value of 36.
So the _hyst_ value is infact 1/2 of the range, while the _over_ is the other half.

Mystery solved. Not too mysterious after all.