When executing a live migration, the vMotion task fails and hits a timeout threshold. The vmkernel.all logs on the source ESXi hosts explicitly record dropped packets and frame size boundary violations on the active physical uplinks:
Host: ####### (Source)
####-##-##T##:##:##.###Z cpu##:#######)Uplink: ####: vmnic#: TSO packet has large MSS (####) plus L# offset (##), which can't be fit into MTU (1500)
####-##-##T##:##:##.###Z cpu##:#######)Uplink: ####: vmnic#: TSO packet has large MSS (####) plus L# offset (##), which can't be fit into MTU (1500)
####-##-##T##:##:##.###Z cpu##:#######)Uplink: ####: vmnic#: Non TSO L# payload size exceeds uplink MTU. FrameLen: ####, L# header offset: ##
####-##-##T##:##:##.###Z cpu##:#######)Uplink: ####: vmnic#: Non TSO L# payload size exceeds uplink MTU. FrameLen: ####, L# header offset: ##
####-##-##T##:##:##.###Z cpu##:#######)Uplink: ####: vmnic#: Non TSO L# payload size exceeds uplink MTU. FrameLen: ####, L# header offset: ##TCP: 5.0
vCenter: 7.0 U3t
ESXi: 7.0 U3s
An MTU mismatch exists across the vMotion datapath. The vMotion vmkernel interface is configured to transmit jumbo frames at MTU 9000, but the backing virtual distributed dwitch (vDS) or physical uplink infrastructure is configured with a standard MTU of 1500, causing jumbo frames to be dropped at the interface level.
Steps:
###-DVS-###-###-#).9000.