Australian cloud for business software — built and run in Sydney since 2002

NetLeverage
NetLeverage

vmkping -d -s 8972 fails: find the jumbo-frame MTU mismatch on ESXi storage or vMotion

Technical article
VMware

About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.

When vmkping -d -s 8972 fails but a normal ping works, at least one device between the two ends does not pass 9000-byte frames. Broadcom KB 313061 says that jumbo frames set on an ESXi vmkernel adapter "might be dropped or fragmented" by any device set to pass standard frames only, so every device along the path must allow jumbo frames. The job is to find the lowest MTU on the path, correct it, and test again.

What the symptom means

KB 313061, "Testing jumbo frame pings over the network - VMware ESXi" (Updated On: 09-15-2026, ESXi 7.0.x, 8.0.x and 9.0.x), explains that vMotion and storage traffic gain throughput with an MTU of 9000, called jumbo frames, while 1500 is a standard frame. It lists symptoms of a mismatch: vMotion failures, TLS traffic that "may not be seen on the host" (the TCP handshake happens but "no TLS traffic is seen"), ESXi hosts unavailable for vSphere HA purposes, errors on NFS, FCoE or software iSCSI storage, and vSAN MTU warnings in Skyline Health.

Devices along the path can be virtual switch ports, physical switch ports, switches, routers, vmks and vmnics of other ESXi hosts. The KB adds that a higher frame size on physical hardware than on an ESXi vmkernel adapter does not cause a problem, because standard frame packets can always pass through jumbo-configured devices. The reverse does.

Some of these services send frames with the do-not-fragment bit set. KB 419088 (Updated On: 03-29-2026) says vMotion, vSAN and iSCSI storage do so and honour the MTU of the virtual switch, and that if the virtual switch MTU is set lower than the MTU at an uplink or intermediate switch, traffic for services such as ESXi storage may be dropped, which can result in an outage.

How the test numbers work

KB 419088 states the ping size as the frame size minus 28 bytes: "for the standard MTU of 1500 the size should be 1472 and for the jumbo frame MTU of 9000 the size should be 8972". The arithmetic is 1500 - 28 = 1472 and 9000 - 28 = 8972. KB 313061 adds a third test, 1473, which is 1501 - 28, to see whether the physical upstream switch port is configured for jumbo frames. If the 1473 test fails, the KB says the physical switch MTU is likely still at the 1500 default.

The -d option in vmkping means "do not fragment" (KB 344313), so an oversized packet is dropped and not split.

Likely causes

Cause What you see How to confirm
Physical switch port or trunk still at 1500 1472 works; 1473 and 8972 get no reply The 1473 test fails (KB 313061)
vSwitch MTU lower than the vmk sendto() failed (Message too long), the output KB 419088 shows when the virtual switch MTU is exceeded Compare esxcfg-vswitch -l and esxcfg-vmknic -l
vmk MTU left at 1500 The 8972 test fails esxcfg-vmknic -l shows MTU 1500 for the vmk
Far end (other host or storage port) at 1500 Test works to one host, fails to another Repeat from the other end
Test aimed at the local vmk IP Test succeeds but storage still fails Target was not the remote IP (KB 395551)
Wrong network stack tested vMotion fails although a ping on another stack passes Add -S vmotion to test the vMotion stack (KB 313061)

Prove the cause before changing anything

Connect to the host by SSH. The commands below only read.

1. Read the MTU settings. KB 313061 lists three commands:

esxcfg-vswitch -l
esxcfg-vmknic -l
localcli network nic list

KB 395551 uses esxcli network nic list for the same purpose; use that if localcli is not available to you. Note the MTU for the standard vSwitch, each vmk and each vmnic. Repeat on the other host.

2. Test with the remote address. KB 395551 warns that pinging the vmk's own IP "will provide false positive results" because the packet never leaves the ESXi stack, and shows the form vmkping -I vmk <iscsi target ip> -d -s 8972. For example, vmkping -I vmk1 192.0.2.50 -d -s 8972, with vmk1 the vmk under test and 192.0.2.50 the remote storage target. For vMotion, KB 313061 adds the stack: vmkping -I vmk2 -S vmotion 192.0.2.12 -d -s 8972.

3. Walk the three sizes.

vmkping -I vmk1 192.0.2.50 -d -s 1472
vmkping -I vmk1 192.0.2.50 -d -s 1473
vmkping -I vmk1 192.0.2.50 -d -s 8972

How to read the results (KB 419088 and KB 313061):

  • Replies at 8972 mean the path passes jumbo frames.
  • "100% packet loss" at 8972 may mean a device along the path does not support jumbo frames, or that the target is not replying to ping (KB 419088's wording: jumbo frames "may not be supported on a device along the network path, or the target device is not replying to ping requests").
  • sendto() failed (Message too long): KB 419088 says this output "indicates that the MTU of the virtual switch was exceeded".
  • Success at 1472 with failure at 1473 suggests the physical switch port is at 1500. If 1473 works but 8972 fails, KB 313061 says the switch may be set to 9000 without allowing for overhead, and the right overhead varies by switch type, so check the switch vendor.

4. Check it is a VLAN network. KB 313061 says to first check that the connectivity concern is over VLAN connections and not VXLAN or GENEVE.

Fix

Correct the lowest MTU on the path. The KB 419088 order is: configure the physical switches and every switch along the path to match or exceed the target MTU first, then change the virtual switch.

Physical switch ports and far end

Ask the network owner to set the port, trunk or LAG to a value that allows 9000-byte frames plus its own overhead, using the switch vendor's documentation. The same applies to the storage array port or the other host's vmk.

Standard vSwitch and vmk

Risk: a changed MTU can interrupt storage sessions and vMotion that use that switch or vmk, and a virtual switch set lower than the path can cause the storage drops described above. If the switch or vmk carries management traffic (a management vmk such as vmk0 is often on vSwitch0), the change can also drop the management network and the SSH session in which you type the esxcfg-vswitch and esxcfg-vmknic commands below, which the back-out also needs. Make the change from the host console (DCUI or an out-of-band console), or leave management vmks alone. Work in a maintenance window, one host at a time. Save first: copy the output of esxcfg-vswitch -l and esxcfg-vmknic -l off the host (paste it into your ticket or notes), so the old MTUs are on record. Back out: run esxcfg-vmknic and esxcfg-vswitch again with the old value, for example esxcfg-vmknic -m 1500 vmk1 first and esxcfg-vswitch -m 1500 vSwitch1 second, so that the switch is never below a vmk's MTU. KB 419088 gives an order only for the switch against the uplink and intermediate switches, and none for vmks; the same order is used here for vmks so that the switch is never below a vmk's MTU. To raise the MTU, run the switch command first and the vmk command second. Then repeat the tests.

KB 313061 gives commands for a standard switch only:

esxcfg-vswitch -m 9000 vSwitch1
esxcfg-vmknic -m 9000 vmk1

The KB's form is -m followed by the MTU and then the switch name or vmk. The vmnics inherit the MTU from the switch.

Distributed switches

KB 419088 says changing the MTU on a vSphere Distributed Switch brings the uplinks down and up, causing a brief network interruption, and applies to all hosts using the switch. For distributed switches, KB 313061 points to Broadcom's "What are Jumbo Frames" page.

Risk: the change reaches every host on the switch. Save first: record the switch's current MTU in the vSphere Client. Back out: set the recorded MTU again and expect the same brief interruption.

Example (illustrative): vMotion between esx01 (vmk2, 192.0.2.11) and esx02 (vmk2, 192.0.2.12) fails. Both vmks and vSwitch1 show MTU 9000. vmkping -I vmk2 -S vmotion 192.0.2.12 -d -s 1472 replies, -s 1473 shows 100% packet loss, so the first physical switch port is still at 1500. The network team sets the access and trunk ports to carry jumbo frames and the -s 8972 test now replies. vMotion works.

Confirm the fix

  1. vmkping at 8972 with -d replies from the remote address, in both directions.
  2. A real task works: vMotion completes, or the datastore mounts and stays connected.
  3. Skyline Health shows no MTU warnings (for vSAN).
  4. Repeat the test from every vmk that needs jumbo frames to every peer; KB 313061 says to test all other hosts in the cluster for vMotion.

What to watch afterwards

  • Record the MTU for every vmk, switch and storage port in one table, and update it when equipment changes.
  • Test jumbo frames again after switch firmware updates and new host builds.
  • Keep management vmks at 1500 if that is your design; the KB notes that some vmks use standard frames while others use jumbo frames.

Risks and limits

  • These commands have not been run or lab-tested.
  • KB 313061's change commands apply to standard switches only.
  • KB 313061 says the overhead differs for different physical switch types; ask the switch vendor for the figure.
  • localcli and esxcli both appear in the KBs for listing NICs. Use the one your host build supports.
  • NSX overlays (VXLAN, GENEVE) are not covered: KB 313061 says to confirm first that the problem is over VLAN connections.

Sources