Hi, We are using OpenVswitch to connect virtual machines running on KVM hypervisors (Fedora 13, openvswitch 1.0.1). Each hypervisor has its ovs-vswitchd daemon to connect differents VMs on this host. VMs are using VirtIO as network driver.
Communications between VMs on differents hypervisors pass thru an openvswitchd daemon on what we call a "communication server". Communication between VMs and the real word pass thru this "communication server" too. vswitchd on hypervisors are connected to the vswitchd on the "communication server" via GRE tunnels. Each ovs-vswitchd has it ovsdb configuration deamon running on this "communication server". That's how our platform was designed. We have 12 hypervisors, for about 100 VMs running now. The "communication server" is hosting the 12 ovsdb + the ovsdb for the central ovs-vswitchd. All those servers are linked with 10Gbps fiber switch. Everythings works fine, connectivity is ok, VMs on the same vlan can ping each others even if they are on differents hypervisors. They can also ping real servers on the same vlan outside etc. That was our goal so we are quite happy. But times to times, we get very very slow traffic and we can't find where the problem is. Files transfers that usually take 2 or 3 minutes can take more than 60 minutes to complete... after a few minutes, everything seems to get back in order by itself (we don't do anything for that). Can you point us some good tips or direction to diagnose openvswitch? If the "communication server" is heavy loaded (it's not the case on our platform), can it slow down communication even between VMs on the same hypervisor? (due to slow response from ovsdb?) Max traffic bandwith betweens VMs is about 20MBps, that is seems ok to you? We are just begining to dig our problem, any help would be greatly appreciated ;) Thanks Best regards, -- Edouard Bourguignon _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss_openvswitch.org
