Impossibly high network traffic reported

[Kenneth Porter]

--On Monday, November 10, 2008 11:16 AM +0000 Stephen Childs 
<childss at cs.tcd.ie> wrote:

> We're noticing a problem with network traffic monitoring using Ganglia
> (and also Lemon) on our PowerEdge 1950s running Scientific Linux 4
> (~=RHEL4). Every now and then a node will report impossibly high network
> traffic (e.g. 10GB/s or more on a node with 1GBit cards). This throws out
> all the graphs for the cluster. Has anyone else seen this?

How do Ganglia and Lemon measure traffic? I'm using Cacti with an iptables 
counter query on one system that seems to give reasonable data. I've also 
used ntop, which I believe hooks into the tcpdump packet filter machinery.

(I'm offline right now due to my Comcraptic service provider, or I'd go 
google Ganglia and Lemon. So I'm queuing this reply into sendmail until my 
connection goes up again.)