I'm reading the Design Guide - NSX-T 3.1 Multi-Locations-v1.1.pdf and I see that an RTEP MTU greater or equal at 1700 is suggested.
L2 Stretched communication is:
1) From ESXi to Edge Node
2) Edge Node to Edge Node (using RTEP*)
3) Edge Node to ESXi
In point 1) there is a GENEVE incapsulation and decapsulation?
Or the GENEVE traffic incapsulated by the first TEP is incapsulated a second time by the RTEP?
My simple question is: why MTU>=1700 and not MTU>=1600 for the RTEP?
So basically as the Geneve header can be variable, they request MTU 1700 for future use case purposes where they cannot only encapsulate Ethernet frames but another ones such as MPLS for example.
And yes, when a packet goes from the LIF DR located into the ESXi to the DR of the Edge node it goes encapsulated but is not re-encapsulated. However now if you use MTU 1600 is going to work for sure.