[ixpmanager] Spikes on MRTG graphs

[Nick Hilliard (INEX)]

Andrei Dinu wrote on 28/03/2020 13:06:
> We also face these issues at Interlan IX. From time to time we normalize 
> the data using https://github.com/mhagander/rrdclean .
> 
> The tool analyzes the values inside the RRD database and you can select 
> a cutoff point. All values above the cutoff point are modified to the 
> value specified.

automated tools can help for RRD, but the underlying problem relates to 
the difficulty in accurately collecting counter data from networking 
devices. Counter overflows and underflows are a fact of life and it's 
hard for third party packages like MRTG to handle situations where snmp 
returns values which are obviously wrong from a semantic point of view, 
but syntactically correct.  There is some sanity checking in there, but 
the more ports deployed, the more likely you are to see weird entries in 
there.

Nick