id: PG-12
category: networking
title: Ping works, big transfers stall?
situation: >-
  Instances of one compute answer ping and accept connections, but SSH hangs on a large output and HTTP
  times out. Agents alive, instances `ACTIVE`, nothing in the logs.

diagram:
  type: compare
  caption: Same instance, same path, two packet sizes
  columns:
    - kicker: Small packet
      title: "`ping -M do -s 56`"
      tone: ok
      points:
        - Still fits the tunnel once encapsulated
        - Most monitoring stays green
      verdict: Answered
    - kicker: Largest packet of the network
      title: "`ping -M do -s <MTU - 28>`"
      tone: stop
      points:
        - Too big for the tunnel interface of that compute
        - Dropped without any error
      verdict: 100% loss
  note: >-
    The tunnel interface MTU must be at least the network MTU plus 50 bytes (VXLAN over IPv4; 70 over
    IPv6).

checks:
  - label: MTU Neutron gives the network
    hint: openstack network show <network> -c mtu
  - label: Don't-fragment ping, two sizes
    hint: "from the router namespace: ping -M do -s <mtu-28> <ip>"
  - label: Tunnel interface MTU, per compute
    hint: ip link show <tunnel-interface>
  - label: The underlay between hosts
    hint: ping -M do -s <tunnel MTU-28> -I <local-ip> <remote-local-ip>

caution:
  - level: state-changing
    text: >-
      Changing the MTU of an interface that also carries management traffic can cut it. Use the out-of-band
      console if in doubt.
  - level: state-changing
    text: >-
      Lowering `global_physnet_mtu` affects new networks only: existing ones need `network set --mtu`.

lesson: A green ping proves small packets pass. Test the largest one the network promises.

runbook:
  path: networking/tunnel-mtu-mismatch-large-packets-dropped.md
  status: lab-verified
  release: "2026.1"

scope: >-
  Lab-verified on OpenStack 2026.1 with Kolla-Ansible and ML2/OVS, VXLAN over an IPv4 underlay, for a tunnel
  interface whose MTU was lowered on one compute. A switch on the path dropping large frames, a platform-wide
  mismatch and an IPv6 underlay were not reproduced. Neutron never re-reads the MTU of the hosts after
  deployment. Instances pick up a new MTU at DHCP renewal or reboot. The global_physnet_mtu change was not run on the lab. OVN was not tested.

linkedin:
  title: "Ping works, large transfers stall: test the largest packet"
  caption: |
    Users say the service is "slow" or "hangs". Ping answers. SSH logs in, then freezes on a large output. Agents are alive, instances ACTIVE, the logs are clean.

    On one compute, the interface that carries the VXLAN tunnels has a smaller MTU than Neutron assumes. Small packets still fit once encapsulated. Full-size packets do not, and are dropped without any error.

    One test shows it, a don't-fragment ping at two sizes:
    → ping -M do -s 56: answered
    → ping -M do -s <network MTU - 28>: 100% loss

    The rule: the tunnel interface MTU must be at least the network MTU plus 50 bytes for VXLAN over IPv4.

    A green ping only proves that small packets pass.
  hashtags: [OpenStack, Neutron, Networking, MTU, VXLAN, OzarLand]
