Hi Aaron or Maxime, 

I want to get your perspective on our testing goals for virtio in DTS if you are willing.

Dean, who works on DTS, has been running various PVP and PVVP virtio workloads on one of the SUT servers at UNH, as well as some single VM and double VM (with inter VM DPDK forwarding) virtio workloads just on his laptop, so that we can start to get an idea of how we can start validating DPDK virtio.

We are aware that there is likely a need/desire for us to setup some vhost-user + virtio testsuites, i.e. there is a tester (traffic gen) server paired up against a SUT server, the SUT server sets up vhost-user device(s), creates a VM(s), and creates a virtio-net-pci device which is connected to the vhost socket from the host, and then we start testpmd in the VM using the virtio-net-pci device(s). Then we send traffic from the TG and assess the DPDK behavior inside the virtio VM. Or, we can run the testpmd inside the VM frontend in tx_only mode and transmit traffic to the backend vhost testpmd. In any case, these are vhost testplan specific details which don't really pertain to my real question which is below.

The question I have for you is do you think there would be any benefit to producing testcases that validating DPDK usage of virtio-net-pci devices in a VM, but without involving Vhost, and keeping the entire "test" constrained to a single host. By this I mean, instead of setting up some vhost/testpmd application on the SUT and forwarding packets from physical NIC ports to the vhost vdev (which is then accessed by the VM virtio interfaces), the testcases would involve creating a TAP interface(s) on the SUT, then making a VM and accessing the TAP interfaces via a virtio interface in the VM, and then just starting Scapy on the SUT host and sending traffic at the TAP interfaces, which goes into the VM? This would mean that the test would not involve any physical devices, but it would still be validating DPDK Virtio. One reason why this is attractive is that it would not require particular hardware, I.e. it runs on 1 system and that system could even be your laptop since it only relies on virtual interfaces. David had asked about this possibility at Prague, and I think at that time Maxime piped up to say "and this would also be useful for virtio" or something like that. Anyhow, let me know if you think this makes sense or any other thoughts you may have. If it is a reasonable direction to go in we can start drawing up test plans.

I realize this is kind of a wall of text... happy to discuss at the CI meeting on Thursday if that is better.

Thanks,

Patrick