The following diagram depicts a series of nested encapsulations representing a packet from a pod in a Kubernetes cluster that is VXLAN encapsulated for transport to a pod on another node (both nodes are VMs in a hypervisor-based compute cluster.)  The same packet is then GENEVE encapsulated, when the packet is routed between hosts in the compute cluster.  It is then VXLAN encapsulated again as the packet is routed across the data-center EVPN fabric.   Each of those additional encapsulations reduces the maximum viable MTU value of the original Kubernetes pod that generated the initial packet.  In this illustration, an original IP datagram of 8500 bytes, ends up as an 8776 byte Ethernet frame after the multiple encapsulations.

image