In the Nodes (Detailed) dashboard, a couple of graphs show statistics about the network interfaces found on the node. This includes stats about the cali* (veth) interfaces, which clutters the views a lot, and doesn't really have much value (IMHO), since one can't map the interface name to some Pod name anyway.

I suggest to

• either remove these interfaces from the Prometheus query (device!~"cali.*", if I'm not mistaken)
• or, remove these interfaces from the Prometheus query, but add a 'virtual' interface in the same panels which aggregates all cali.* interface traffic (which would allow to discern (most) container-based traffic from host-initiated traffic).