[collectd] Hit counter in threshold

Mariusz Gronczewski xani666 at gmail.com
Sun Jul 12 10:01:54 CEST 2009


Hi,

Generally hysteresis have wide meaning, for example in typical
electrical circuits by hysteresis they mean something like:
a)go from LOW to HIGH when above 4V
b)go from HIGH to LOW when below 2V
so u have 2V "noise windows"

tho both definitions are useful in practice (for example above one
would be good for temp, "turn off alarm when its 5 degrees below MAX",
and "trigger if 3 samples over" would be great for CPU or load avg
meter)
So something like
<Match "value">
  Min   0
  Max 100
  Hits 3
  Hysteresis 5
  Invert true
  Satisfy "Any"
</Match>
(match if 3 samples are below 0 or above 100 and stop matching if
value is between 5 and 95)
would probably satisfy all needs ;]

2009/7/12 Peter Bray <pdb_ml at yahoo.com.au>:
> Hi,
>
>>  1) hysteresis, i. e. something like this:
>>     -- 8< --
>>       if (value > (threshold + epsilon))
>>         status = bad
>>       else if (value < (threshold - epsilon))
>>         status = good
>>       return status
>>     -- >8 --
>>     This, used right, should provide modest flapping protection.
>
> Just a small comment, hysteresis as used in the satellite control
> systems I've worked with over the last eight years, means N samples
> above/below a threshold to trigger the condition. So for example: if RH
> (Red High) >= 10.0 with hysteresis of 3, then you require at least three
> samples at or above 10.0 to trigger a red high alarm. This also seems to
> match the definition in the MacOSX Desktop Dictionary "the phenomenon in
> which the value of a physical property lags behind changes in the effect
> causing it". And yes we used this a flapping protection on noisy data
> from the satellites (Solar Wing Position Potentiometers come to mind,
> where we could experience HUGE 1 sample spikes in the data)
>
> Others may have a different understanding of hysteresis but I thought I
> mention this understanding as naming is so important to understanding.
>
> Regards,
> Peter
>
> _______________________________________________
> collectd mailing list
> collectd at verplant.org
> http://mailman.verplant.org/listinfo/collectd
>



More information about the collectd mailing list