> Should we try to restard ovs-vswitchd daemon? Because it's working > fine on the other VMs.
We try to restart the ovs-vswitchd with no change. We also try to switch from virtio net driver to e1000 driver same problem. Each time the VM is poweroff/destroy, its tap devices are removed from ovs and then recreated with same port name at vm creation. Restarting/destroying the VM doesn't solve the problem. The problem is hard to diagnose because it's not static, we can't reproduce it at will. When a port is removed, and then a new one is created but with the same name, does it "inherit" the old datapath of the previous port? The slowness/packet loss seems to occur when the hypervisor is having some IO stress. We don't have the problem on the first couple of VM we install on the host. It occurs more often when we already have few vms which are using the network thru ovs. The ratio between hit and missed values in ovs-dpctl is the same in both case (normal and slow network). The slowness only impact some port on the vswitched. Other VMs on the same vswitchd keep working fine. So If anyone have any idea, it should be greatly appreciated. By the way, there's still no tar.gz of openvswitch 1.0.3 on ovs website. We'd like to test this version, can we upgrade safely from 1.0.1 (no change in the database format?). Any particular procedure? -- Edouard Bourguignon _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss_openvswitch.org
