Here are the results of my vNIC performance testing.  These tests were conducted on a 9119 MHE (E880).  No other workloads were running at the time of the testing so these results should provide a very good indication of achievable throughput when using vNIC.

We have the EN0J PCIe3 4-Port 10Gb FCoE & 1GbE Cards.  Each test was run eight times to get a good average set of results.

For these tests, the VIO Servers had a SEA adapter configured from the same SRIOV Adapters and they were bridging the same VLANs.  So extra VIO Server CPU utilitsation is reported due to the seaproc threads inspecting each packet to see if it needs bridging or not.  I conducted exactly the same tests without the VIO Server SEA adapters to show only the VIO Server CPU utilisation for vNIC processing. 

For reference, here are my LPAR and VIO Server Configurations.

1 Thread Test

Client command: /usr/bin/iperf -c 192.168.30.181 -fm -P1 -l1M   -t60

8 Threads Test

Client command: /usr/bin/iperf -c 192.168.30.181 -fm -P8 -l1M   -t60

10 Sessions Maximum Transactions per Second

Client command: /usr/bin/netperf -H 192.168.30.181 -l60 -v50 -t TCP_RR -- -r 700 -D

50 Sessions Maximum Transactions per Second

Client command: /usr/bin/netperf -H 192.168.30.181 -l60 -v50 -t TCP_RR -- -r 700 -D