I'm wondering what maximum performance numbers others are seeing with Workstation Pro.
I'm running against a wall at about 2.8 Gb/s with iperf3 when sending and 230 MB/s smb file copy when the VM is the source.
When the VM is the iperf3 receiver it's about 5.6 Gb/s and for copying files from a network share about 390 MB/s with the VM as the target.
I did run my tests on a 10GigE setup which delivers about 9.8 Gb/s with iperf3 on the VM host in both directions and about 1.1 GB/s when copying files between the VM host and the smb servers.
I tested with Workstation Pro 15.5.7 using the vmxnet3 network adapter with the latest driver version 1.8.17 and RAM disks as storage on the VM side.
It looks like the limiting factor is related to the interaction with the kernel. Altering various driver parameters like ring buffer sizes, number of queues and interrupt parameters didn't result in remarkable changes. When running the tests inside the VM you can see the kernel occupying one core to almost 100% all the time. When running on the host the kernel CPU load is almost negligible.
While this could somehow explain the root cause of the overall performance limitation I do not understand the big difference between sender and receiver mode. I was surprised that there is not much information on this issue and only found a few related threads.
However, today 10GigE networking has made it into small business and home environments with more and more affordable hardware entering the market. I would also like to understand if VMware is aware that this limitation is becoming an issue for the desktop virtualization and if there are plans to do something about it.
The intention of my post is not to discuss my particular setup but as stated in the question to understand what performance numbers others are seeing over 10GigE from inside a VM and of course if anybody out there was able to at least get close to the theoretical maximum performance.
Before going into the details of the setup I really would like to understand if it's possible at all.
So did you perform any 10GigE performance tests inside a VM and can share your numbers ?