Test Results

The performance of OVS flow table normalization is improved by more than 50%.

These tests did not test the performance of forwarding cores directly. This is because the performance cannot be tested directly. Moreover, there is no need to do so because the test environment is similar to the production environment, and thus the tested performance is similar to the actual performance.

The number of network port queues and VM queues affects the performance. The number of queues in the tests is twice the number of forwarding cores.

The performance of the OVS flow table normalization solution is better than that of the native OVS+DPDK, and more VMs are required to test the pressure of forwarding cores. In the VXLAN+CT scenario, multiple VMs and forwarding cores decrease the performance of the native OVS+DPDK due to locks in the CT. On the contrary, the OVS flow table normalization solution bypasses some locks, and the performance is greatly improved. Therefore, the OVS flow table normalization solution works better than the native OVS+DPDK when multiple forwarding cores and VMs are involved.