Wrong Data in some Graphs - comparision with mrtg

[Sirup]

I have Cacti 0.87e on a FreeBSD Machina 7.2.
For a specific device I think the polled data is wrong.
Template is SNMP - Interface Statistics.

Parallel to Cacti I have MRTG running on the same machine. MRTG was installed before and should be replaced though Cacti.
All graphs are identical between MRTG and Cacti.
Only one device shows huge differences.

This only happens to 3 graphes - the other ~15 show the same results.

Screenshots should explain.

[BSOD2600]

Are you positive graph 1 and 2 are polling the same OIDs? They don't even look remotely simular.

[Sirup]

100% sure - Its just interface statistics - same port.

But the look the same - the difference is cacti has a spike at 20 und 22 about 50m.

[BSOD2600]

I'm not entirely sure they are... for example you're middle set of graphs, the inbound/outbound sources are switched between cacti and mrtg.

You've actually compared the mrtg config with the poller cache for OIDs used?

Anyways if they are the same, the only things I can think of would be interface reset / up / down, etc. Maybe gandalf knows more details how rrdtool and mrtg differ under the covers when it comes to storing data/spikes.

[gandalf]

Maybe gandalf knows more details how rrdtool and mrtg differ under the covers when it comes to storing data/spikes.

Unfortunately, no, I don't. When using MRTG with rrdtool, both should do exactly the same (spike "killing" etc. is a feature of rrdtool, neither of MRTG nor Cacti (in Cacti default configuration))
R.