Hardware-passthrough (in vmware terminology) is a subset of what it can do, and you can take advantage of SR-IOV without using hardware passthrough explicitly.Īs with all other things in the datacenter, YMMV and you should thoroughly test your applications both in isolation and simultaneously, to see if or how much you benefit or get hurt by it.
SR-IOV is to IO device virtualization what VTx is to CPU virtualization (sorta). That sounds like a pretty large over-generalization. I'm interested in understanding network performance of guests between each other in particular, when SR-IOV is utilized in hyper-v specifically (which is why I am starting here and not in the networking forum).
So, does anyone have any experience to this and can enlighten be a bit about how it is in practice. If I move to SR-IOV, it almost sounded like I wouldn't necessarily have a vSwitch, which I think would be actually a step backward in a lot of ways. The software switch has the advantage in my mind of not having to actually go over the physical network for traffic between guests on the same vSwitch, so performance when accessing the file server (which is a lot) is faster then going out the nic and back in. My concern here is that one of the guests acts in a file server role. I didn't think that was a thing under hyper-V, I thought you had to still always attach to a v-switch and then attach guest vNics to the vSwitch. I was some stuff about how SR-IOV presents, and it sounded like a use case here was having it present hardware directly to guests, thus bypassing the software switch. So I am updating the hardware in my lab VM host to a modern 4-port NIC that supports SR-IOV, a technology I've been aware of, but not really had an opportunity to play with previously.
0 kommentar(er)
