HI Russel, Please find the work flow as below. 1. A user requests nova to boot a vm 2. Nova schedules the vm based on the flavor and image properties to a host 3. Nova asks neutron to bind the neutron port basing the selected host name and some limited information in binding profile. 4. The neutron network driver(ML2/OVN driver) that manages the selected node will then bind the port and select the vif type to use. 5. Nova reads the vif type selected by neutron and generates the Libvirt xml accordingly and boots the vm. a. For the normal interface nova plug the vhost-net interface to the OVS. b. In the dpdk case during the boot process nova plug the vhost user interface into ovs
The VIF type(kernel vhosts/vhost-user/sriov) must be decided at step 4 based on the capabilities of specific compute node to allow nova to generate the correct Libvirt xml. Nova doesn’t know the capabilities of selected compute node to decide what VIF type to be used in the libvirt XML. It’s responsibility of Neutron to provide this information because Nova is not supposed to have any logic for manage the network. Neutron can talk only to the OVN NorthDB and this patch series will expose the relevant compute node details all the way up-to Northbound_DB. So Neutron can then instruct Nova to boot VM with right VIF type using the exposed information in the Northbound DB. Since OVN is taking care of entire cloud network management, we feel that its responsibility of OVN to provide enough information to Nova->Hypervisor for booting the VM with right parameter set. Regards _Sugesh -----Original Message----- From: Russell Bryant [mailto:rbry...@redhat.com] Sent: Tuesday, August 25, 2015 8:20 PM To: Chandran, Sugesh; 'Ben Pfaff' Cc: dev@openvswitch.org Subject: Re: [ovs-dev] [PATCH 0/4] ovn: Schema modification for DPDK/userspace tunneling support in OVN-Openstack integration. On 08/25/2015 06:16 AM, Chandran, Sugesh wrote: > Hi Ben, > > This patchset is for enabling Openstack to deploy a VM on a compute > node with right parameters. Currently Openstack-Neutron cannot get the > compute node information such as datapath type(system/netdev) and > supported port types from OVN. The proposed schema changes is for > exposing these information to Neutron so that it knows the networking > capabilities of each node and Nova could start the VM on that node > accordingly. ODL is also exposing these information to the Neutron for > the same purpose. I don't think OVN will care. I get that OpenStack needs to care. It's not clear to me why the information has to flow through so many layers, though. It seems like this could be handled entirely in Nova. Do you have a good reference that documents the expected workflow in more detail? -- Russell Bryant _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
