I can't imagine that jumbo frames would have much impact on vMotion performance. Were you having bad performance with vMotion previously? With ESX 3.02 without jumbo frames I can vMotion a 4GB virtual machine in about 50 seconds. The theoretical maximum speed for transferring 4GB on a 1Gb link is about 120MB/s and you're unlikely to see more than about 100MB/s in the real world, jumbo frames or not. That means in the absolute best case it would take 35 seconds just to transfer the data and that does not include the setup time, quiesce time, etc so perhaps, assuming jumbo frames helped the throughput, I might be able to get the vMotion of my 4GB system done in 40 seconds rather than 50. Based on other factors it's hard to know if you would even get that much difference.
I think it's far more likely that you would see benefit from jumbo frames if you have virtual machines which make heavy use of network bandwidth. For example, we're seeing some improvement from VM's that use software iSCSI initiators within the virtual machine, not a huge boost in overall throughput, but a minor boost while actually lowering the overall CPU usage of the host which is still a nice win overall. I could also see jumbo frames helping on 10Gb links by lowering the the number of interrups which should lower the overall CPU impact of such high speeds.
Later,
Tom
