Comment by Nodir for Hi I just installed Kilo, Running on RHEL 7 hypervisors, I ran up two VMs on different hosts with a 10G link between them. With SR-IOV I get great performance as expected. With standard virtio + vhost, the performance is very poor. A iperf test, shows I can get line speed between 2 VMs (when using virtio) iperf -c 192.168.0.5 ------------------------------------------------------------ Client connecting to 192.168.0.5, TCP port 5001 TCP window size: 22.5 KByte (default) ------------------------------------------------------------ [ 3] local 192.168.0.10 port 57195 connected with 192.168.0.5 port 5001 [ ID] Interval Transfer Bandwidth [ 3] 0.0-10.0 sec 8.78 GBytes 7.54 Gbits/sec But when running a load test for a http websever, at around 2000 http requests, the throughput appears to slow (latency increases) although packets aren't getting dropped on the VM interface. N.B that I don't see this issue with SR-IOV, which makes me thing the VM settings are fine, but it could be the virtio driver, hypervisor config (possible ovs) or host nic driver. The tap devices have a low txqueue length, but this is the same for the last release. Is there any default throttling going on with the latest version of Kilo. Can anyone recommend tunings for the hypervisor or virtio driver to improve the performance? In past versions of Openstack I've been able to hit around 5000 http requests per second.

> Is there any default throttling going on with the latest version of Kilo. Did you see packet drops on VM's tap interface? If you did not check it, can you log in to the host where VM is running and do "ifconfig tap-xyz" to see if packet drop counters increase?