vNIC MTU behaviour

[pieman2015]

If we have a server profile that has a vNIC profile with the MTU set at 9000 (with no associated COS policy or the upstream switch having large frame support enabled), what is the behavior of traffic flowing out of the host? 

Given UCS sets the driver automatically in the OS, would say jumbo frames be enabled in Windows and thus cause fragmentation issues?

[Justin Welch]

Hello,

If jumbo frames are configured in the OS then yes this can potentially cause tail drops or fragmentation before the traffic can make it from the FIs. Keep in mind that MTU is maximum transmit unit so not every packet will be 9000 it depends on the payload.

If you are configuring jumbo frames in an environment you will need to make sure the MTU value for the entire path of the packet is 9000 minimum (9216 recommended in the event of protocol encapsulation).

Thanks,

Justin

[pieman2015]

Thanks for the reply.

I should clarify.

If the UCS profile has 9000 configured as the mtu under the vNIC profile and no where else, what kind of performance issues if any could occur?

I understood that the enic driver new what setting was applied at the ucs level and changed it according to the ucs setting?? If this is the case, wouldn't setting 9000 at the ucs nic profile level cause performance issues?

[Justin Welch]

Hello,

This goes back to what the payload is. In example, I can ping the default gateway and the traffic will make it with no issue because we don't need to use an MTU about 1500 for the transmission of the traffic. If we need to use an MTU larger than 1500 since the opposing end of the link from the NIC does not have an MTU of 9000 the packet will be dropped.

if we have a configuration more like this:

9000 FI 9000 - 9000 5K 1500 - 1500 CAT3750

we would fragment the packets between the 5k and the 3750 because we have to take the 9000 MTU packet, split it into multiple 1500 MTU packets and forward it.

Let me know if this helps.

Justin