My desired architecture includes a Linux VM which runs the controller parts of OpenStack (including all of the API endpoints), one (for now) bare-metal network node (L3, DHCP and maybe metadata) and many bare-metal compute nodes. I'm trying to install thusly: CONTROLLER=10.0.1.10 NETNODE=10.0.2.10 COMPUTENODES=10.0.2.20,10.0.2.21,10.0.2.22,... TENANT_NET_TYPE=vlan VLAN_RANGES=pubphysnet01,priphysnet01:100:199 BRIDGE_MAPPINGS=priphysnet01:br-bond0 BRIDGE_INTERFACES=br-bond0:bond0 packstack \ --neutron-l2-plugin=ml2 \ --neutron-ml2-type-drivers=flat,${TENANT_NET_TYPE} \ --neutron-ml2-tenant-network-types=${TENANT_NET_TYPE} \ --neutron-ml2-mechanism-drivers=openvswitch \ --neutron-ml2-flat-networks=pubphysnet01 \ --neutron-ml2-vlan-ranges="${VLAN_RANGES}" \ --neutron-ovs-bridge-mappings="${BRIDGE_MAPPINGS}" \ --neutron-ovs-bridge-interfaces="${BRIDGE_INTERFACES}" \ --neutron-l3-hosts=${NETNODE} \ --neutron-dhcp-hosts=${NETNODE} \ --neutron-metadata-hosts=${NETNODE} \ --install-hosts=${CONTROLLER},${COMPUTENODES} This presents two problems: 1) (less intrusive, but annoying) neutron-openvswitch-agent is installed and enabled on the controller VM, which I don't want (I can disable the service, 'neutron agent-delete', and unwind the ovs config that it did) 2) neutron-server gets launched on the network node (10.0.2.10) as well as the controller (10.0.1.10). Curiously, the service is not enabled on the network node (chkconfig --list shows 'off'), but it is running when packstack completes. This causes mass confusion for the agents, as there are two neutron-server's running at the same time. It results in obscure problems, including apparent messaging failures, and agents reporting tables missing from the database (neutron.conf on the network node doesn't have the SQL connection info). Am I doing something wrong here? Should it work as above? I can file a bug or two, but thought I'd check here first.... Tnx, ~iain