The Maximum Transmission Unit (MTU) health check, also called "MTU check (ping with large packet size)" complements the basic connectivity check for vMotion. 
Large MTUs, often referred to as jumbo frames, are configured to improve network performance.
Incorrectly configured MTUs might cause performance issues or I/O errors in individual objects. 
Configuration errors can also lead to a failure in virtual machine migration from one physical server to another. 
For high availability of vMotion, it is important for the large ping test check to succeed.

The MTU check uses large packets of up to 9000 Bytes. 
When the small ping test succeeds, the large ping test also succeeds when the MTU size is consistently configured across all VMkernel adapters, virtual switches, physical network adapters, and any physical switches.
The correct configuration assumes that MTU is consistently configured across the vSAN cluster. 
All network components along the network path must use an MTU value that is equal to or higher 
then the value configured on the source network component, but does not exceed the maximum 9000 Bytes value.

For example, you can use the MTU of 1500 on the VMkernel adapter and the MTU 9000 on the physical switch. 
If the source VMkernel adapter has an MTU of 1500, it will fragment the 9000 Byte packet. 
The fragments then travel along the network to the other ESXi host where they are reassembled. 
As long as all network devices along the path use a higher or equal MTU, the test passes.

 

Generally speaking, the Test will fail if there is an MTU misconfiguration and/or malfunction on the vMotion Network. 
Check the configuration/functionality of VMkernel adapters, virtual switches, physical network adapters and any physical switches.
For more information, see Testing VMkernel network connectivity with the vmkping command (1003728).


Following are some Examples of when the Test will fail:

A.) MTU Misconfiguration
The VMkernel adapter on the Host has the MTU of 9000, but the physical switch enforces the MTU of 1500. 
The failure occurs because the source does not fragment the packet and the physical switch drops the packet.

B.) Stretched Clusters
This test regularly fails against the witness nodes in vSAN stretched clusters  - such failures are considered cosmetic only and have no impact on the environment
This is a known issue and will be fixed in a future release.

C.) Network Stack
This check could fail if the vMotion ping failed due to the vmknics using the vMotion network stack. This issue was fixed in vSphere 6.5 Update 1.

D.) Packet loss
This test will also fail if there is packet loss along the link.