On Sun, Dec 21, 2014 at 10:58:17AM +0200, Lior Vernia wrote:
On 19/12/14 22:19, Amador Pahim wrote:
> On 12/19/2014 02:08 PM, Darrell Budic wrote:
>> I tried a quick iperf test a while back and got 3-4Gb/sec between a
>> pair of VMs on separate hosts with a 10G infrastructure, no real
>> tuning and no SRIOV. That met my needs so I didn’t try anything
>> further. If you were aiming for 10G for all, you’d want to work on
>> SRIOV I imagine, but they get pretty good performance even without it.
>
> Maybe we should report vNIC speed as the speed of underlying NIC/Bond.
> Even being the underlying NIC/Bond shared, it would reflect better the
> vNIC max. speed than a fake hard coded value.
>
Or maybe even have it configurable from the GUI?...
... or maybe hide the "speed" as deep as possible, as it is one big lie.
It's a lie when discussing physical nics (where speed is affected by
host load or switch buffers) and it is certainly a lie when it comes to
vNICs, with over-commitment of bandwidth.