On Tue, Apr 05, 2011 at 05:13:59PM +0200, Madko wrote: > We have the same needs about STP (virtualize network as real as it gets), to > simulate network topologies we'd like to use in production. The main > goal is to eradicate any SPOF. So is it possible to attach a VM with a > bonding linked to two differents vswitch linked together to > prevent/simulate failure of one vswitch? This will certainly make a > loop, in real life STP will cope that problem, how about openvswitch? > > For the HA problem, on the physical side, we have a central server > hosting a vswitchd hub linked to other vswitchd using GRE tunnels. > This host is clearly a SPOF. If this host has a major failure, is it > possible to swap/migrate ??to an other (backup/slave) host the IP used > by the other GRE tunnels to contact this vswitch hub? As we do with > other services using keepalive/VIP/VRRP. > What are your recommandations to prevent failure of the vswitchd host? > May be a network topology with a central hub/vswitch is not a so good > idea, but what are the other choices? Is it possible to make a mesh > network without any SPOF with openvswitch? How to interconnect > different hypervisors?
One reasonable way to avoid a single point of failure for physical ports is to use an active-backup bond with slaves connected to different switches. For GRE tunnels, OVS doesn't have an automatic solution. You can use an OpenFlow controller to notice tunnel failures (e.g. via OVS's 802.1ag CFM support) and implement failover at the controller. _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
