[rrd-users] Dealing with spikes due to reachability
cgriffin352 at gmail.com
Wed Apr 25 14:12:59 CEST 2007
I know the subject of spikes is a fairly popular discussion. We have an
issue here where our equipment will become busy for short periods of time
(usually a couple of minutes). We poll for interface information on 1
minute intervals which results in several NaNs. Once the device is again
responsive to SNMP, we usually get a value that is equal to all the traffic
that has passed during the NaN interval. This results in a spike in the
graph. During high usage times, this spike will not occur as it exceeds the
maximum defined value in the RRA, but during low usage we do get the spike.
After researching it seems that its always the value that directly follows
one or more NaNs that is the issue. Is there a way within the graph
function to ignore any CF value that follows a NaN as unreliable as far as
graphing goes (which should get rid of my spike)? It seems that would also
take care of devices which reboot?
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the rrd-users