Overlay VM to VLAN VM Ping Fails for Higher MTU Packets in NSX-T L2 Bridging Environment

Products

• VMware NSX

Issue/Introduction

When attempting to pass traffic across an L2 bridge configured in VMware NSX-T, you observe the following symptoms:

• L2 bridging is configured on NSX with the default gateway (DG) residing on the physical router.
• Pings with small payload sizes from an Overlay VM to a VLAN VM (or the Default Gateway) succeed perfectly.
• Pings with larger payload sizes (Jumbo Frames / higher MTU) fail with 100% packet loss when the "Don't Fragment" (-d) flag is applied.
• Specifically, a ping with a payload size of 1414 bytes succeeds, but a payload size of 1415 bytes or higher drops completely.

To isolate whether the bottleneck is on the Geneve overlay configuration or the physical network layer, execute a TEP-to-TEP ping test directly from the ESXi host to the Edge Node:

• ping ++netstack=vxlan -I vmk10 10.xx.xx.xx -s 1472 -d

Environment

VMware NSX

Cause

This issue occurs because the physical underlay network path directly attached to the ESXi host is hard-capped at a standard 1500 MTU, despite the NSX Edge Uplink Profile and Distributed Virtual Switch (DVS) being correctly configured for Jumbo Frames (MTU 9000).

Resolution

To resolve this issue, the physical network infrastructure must be updated to support Jumbo Frames:

• Check Physical Switch Settings: Coordinate with your network team to review the MTU configuration on the physical switch ports connected to your ESXi hosts (specifically trunk ports handling TEP/Geneve traffic).
• Increase MTU: Ensure the entire end-to-end underlay physical path is configured to support Jumbo Frames (a minimum MTU of 1600 is required for NSX-T overlay traffic, though 9000 is best practice).