We're using vSphere 5.0 U2. We have a SR open with VMware because of some networking issues in our environment. Specifically with Sharepoint.
Our environment is on a DVS with the MTU set to 9000. The virtual machines are on a VLAN that is jumbo enabled on the switches. The TSE at VMware is stating that we should **not** be setting the MTU on our VMs to 9000, but only the DVS. My networking team completely disagrees with this and I tend to agree with them.
We also have many physical servers and storage on this same VLAN whose interfaces do have the MTU set to 9000. Does anyone have any insight to VMware's saying to leave the MTU at 1500?
Thanks,
Clint
Our environment is on a DVS with the MTU set to 9000. The virtual machines are on a VLAN that is jumbo enabled on the switches. The TSE at VMware is stating that we should **not** be setting the MTU on our VMs to 9000, but only the DVS. My networking team completely disagrees with this and I tend to agree with them.
If your goal is to use jumbo frames for guest network traffic, then the TSE is incorrect. Perhaps he or she is thinking that you want to use jumbo frames for the IP storage (NFS / iSCSI)? In that case, the guest network would not need to be modified.
Make sure your guest VM is using a vmxnet3 vNIC. For documentation showing that you can set jumbo frames on the guest, see the KB article below.
Enabling and verifying IOAT and Jumbo frames
Thank you, that is what I thought. We do intend to use jumbo frames for guest network traffic.
The TSE at VMware is stating that we should **not** be setting the MTU on our VMs to 9000, but only the DVS.
Did they provide any reasoning for this?
Within one layer 2 domain you should configure the same Ethernet MTU on all devices End-to-End, that includes VM vNICs, vmkernel ports (if relevant), vSwitches, physical switches, physical servers/storage arrays etc within this layer 2 domain/VLAN.
It should however usually work even with misconfigured MTUs on the end points with TCP connections, due to how the end points "announce" their MTU via TCP MSS clamping during the TCP handshake. At least as long as the layer 2 forwarding components (vSwitches, pSwitches) along the way support forwarding the maximum frame size the end points agreed on.
See:
http://rickardnobel.se/different-jumbo-frames-settings-on-the-same-vlan/
http://blog.ioshints.info/2013/01/tcp-mss-clamping-what-is-it-and-why-do.html
We're still waiting on a reply from the TSE. I'm not directly involved with the support call, but I was told the TSE said he/she's never heard of changing the MTU within the guest, which went against everything I've heard and read.
Thanks for those links, I have a better understanding of why jumbo pings workwithout the propper MTU.
