DC Core and Edge Networking - is there any reason against going full MTU 9000?

Hi all,

I have been doubting myself with this question since knowing about overlay networking (with the use of VXLAN). Apparently, in some networking Leaf-Spine topology, all Leaf and Spine interface MTU seems to be set to more than 9000 (in order to cater for both normal and jumbo frames + VXLAN/UDP/IP overhead). Endpoints on two ends of a connection, with default configuration, send normal frames (1500B or less, caught in Wireshark) and communications seem to happen normally.

In some cases with dynamic routing (OSPF), it requires to set MTU on both sides to the same (because MTU ignore implementation is unstable across vendors), but most modern DC networking devices nowadays also support jumbo frames. So, maybe there's an exception for legacy routing devices?