1470 yes, 8745 no. The reason why is because the witness VM is at a colocation facility therefore, the ROBO office and the main datacenter are connected through the WAN link via VPN tunnel so jumbo frames are not supported. Having said this however, I changed all MTU settings on the 2-node vSAN cluster (vDS, vSS, vmk port groups) and the witness settings to 1500. The results are the same.
The physical switches are set to jumbo frames at the colo and at the ROBO so bringing down the MTU on both sites to 1500 should not represent an issue.
Last night however, I ran the health test multiple times and there was no failure. I have to assume that latency causes the warnings during the day, but the bandwidth requirement for the witness VM is 2.0 Mbps for 1000 components (we are way below that amount of components) and 200ms RTT.
