Zabbix 1.1.4 - Wrong graping?

[tore]

I'm using Zabbix 1.1.4 with Linux agents (also 1.1.4) (latest version in Debian stable).
I've configured input and output statistics (unix_t template) as Numeric(float) Type of Information, Bps Units and Delta (speed per second) as Store Value.
Most of my hosts report sensible values x and y KBps and so on. But 3 of 20 hosts give output that is completely insane, i.e. a delta of +/- ~800 GBps on both net.if.out.eth0 and net.if.out.eth1. This applies to both input and output, but not always both and not always both eth0 and eth1.
Anyone got an idea of what might be wrong?

Regards

[bluelinq]

Similar to this I had a Windows server reporting 20% plus CPU utilization while task manager had it completely flat and this went on for about 1 hour or so. I restarted the agent service and it came back to normal but I also wonder what caused the issue.

Regards,

Paul