* There is high latency when testing latency between VMs in an Openstack environment.
* This affects OVS in P/Q.
* Setup Openstack.
* On a compute node create at least 15 tenants and add 50 instances (more instances show larger latencies).
* From the Neutron node ping to a particular instance:
ip netns exec qrouter-<uuid> ping -c 1 <instance ip>
* We expect this latency to be similar to the latency from not using OVS.
* The following patch from upstream OVS solves the issue: c37d4da4f0fd0190ea8fef97399df6d27ebf6ee2.
* It requires 755be9ea9d7250edd37e1056362e08a7436f6983 to be backported as well since the 1.4.6 still uses sysfs to get carrier info (/sys/class/net/%s/carrier) instead of ioctl. Once we are using ioctls the second patch c37d4da4f0fd0190ea8fef97399df6d27ebf6ee2 reduces the number of times the ioctl is called from netdev_get_carrier().
* There has been testing already done with these two patches applied on top of 1.4.6 in precise.