Jumbo Frames Clarification

[OsburnM]

Greetings— Looking for a clarification on configuration of jumbo frames on a new cluster. Hosts are 2x25GB connected to Cisco switches supporting jumbo frames. Hosts will run esx mgt, vmotion, vsan, & VM traffic all on the same pair of uplinks attached to a vds.

Is it considered best practice to have the vds, vmotion, & vsan interfaces all 9000 but mgt and VM portgroups all still 1500?  Or do the VM portgroups also need to be 9000 since the uplinks are as well? 

[IRIX201110141]

You cant set Jumbo Frame  on VM portgroups.  its the VM Guest which controls its MTU.

Regards,
J.B

[RajeevVCP4]

You are correct (vmotion, & vsan interfaces all 9000 but mgt and VM portgroups all still 1500)

About vm port group , it depends on your work load or use case ,

For further plz check this docs

https://docs.vmware.com/en/VMware-vSphere/7.0/com.vmware.vsphere.networking.doc/GUID-1A028E33-3E41-4...

[OsburnM]

Thanks for the reply-- I did mispeak-- I meant interfaces (vmkernel interfaces or guest interfaces).

I understand the physical switches would be set at 9000 and the vDS itself would be set at 9000 and then the vmkernel interfaces would be set at 9000--  what I'm a little confused on is the guest interfaces.  If everything else is 9000, do the guest interfaces also NEED to be at 9000, or will leaving them at 1500 be okay?  My understanding is it would however if it were reversed (guests at 9000 but switch/uplinks at 1500 would result in a fragmentation problem.

This is new for us-- never entertained jumbo frames before but we were asked by DellEMC (for a VxRail implementation).

[a_p_]

Network packets sizes are defined by the end devices. For software iSCSI that's the ESXi host VMkernel port groups on one side, and the storage system on the other side. All switches/devices between them have to be able to handle the sender's packet size. With Jumbo frames configured (usually 9000 or 9216 on some switches), all network packets with this size, or smaller will be transferred to the destination without the need to fragment them. I usually only set MTU 9000 for iSCSI traffic.

André